Predation Drives Specialized Host Plant Associations In

bioRxiv preprint doi: https://doi.org/10.1101/2020.06.16.150730; this version posted June 17, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.

1 Predation drives specialized host plant associations in preadapted milkweed 2 bugs (Heteroptera: Lygaeinae) 3 4 Georg Petschenka*1, Rayko Halitschke2, Anna Roth3, Sabrina Stiehler3, Linda Tenbusch3, Tobias 5 Züst4, Christoph Hartwig5, Juan Francisco Moreno Gámez6, Robert Trusch7, Jürgen Deckert8, 6 Kateřina Chalušová9, Andreas Vilcinskas3,5, Alice Exnerová9 7 8 *correspondence to: [email protected] 9 10 1Department of Applied Entomology, Institute of Phytomedicine, Faculty of Agricultural 11 Sciences, University of Hohenheim, 70599 Stuttgart, Germany 12 2Department of Molecular Ecology, Max Planck Institute for Chemical Ecology, 07745 Jena, 13 Germany 14 3Institute for Insect Biotechnology, Justus Liebig University Giessen, Heinrich-Buff-Ring 26–32, 15 35392 Giessen, Germany 16 4Institute of Plant Sciences, University of Bern, Altenbergrain 21, CH-3013 Bern, Switzerland 17 5Fraunhofer Institute for Molecular Biology and Applied Ecology (IME), Branch for 18 Bioresources, Ohlebergsweg 12, 35392 Giessen, Germany 19 6Sociedad Andaluza de Entomología, 41702 Dos Hermanas (Sevilla), Spain 20 7State Museum of Natural History Karlsruhe, 76133 Karlsruhe, Germany 21 8Museum für Naturkunde, Leibniz Institute for Evolution and Biodiversity Science, 10115 Berlin, 22 Germany 23 9Department of Zoology, Faculty of Science, Charles University, 12843 Prague, Czech Republic 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40

41 Abstract

42 Host plant specialization across herbivorous insects varies dramatically, but the underlying 43 evolutionary mechanisms are little-known. The milkweed bugs (Heteroptera: Lygaeinae) are 44 ancestrally associated with plants of the Apocynaceae from which they commonly sequester 45 cardiac glycosides for defense, facilitated by resistant Na+/K+-ATPases and adaptations for 46 transport, storage and discharge of toxins. Here, we show that three Lygaeinae species 47 independently colonized four novel non-apocynaceous hosts, convergently producing cardiac 48 glycosides. A fourth species shifted to a new source of toxins by tolerating and sequestering 49 alkaloids from meadow saffron (Colchicum autumnale, Colchicaceae). Across three species 50 tested, feeding on seeds containing toxins did not improve growth, but sequestration mediated 51 protection against predatory lacewing larvae and birds. We conclude that physiological 52 preadaptations and convergent phytochemistry facilitated novel specialized host associations. 53 Therefore, selection by predators on sequestration of defenses, rather than the exploitation of 54 novel dietary resources, can lead to obligatory specialized host associations in generalist insects. 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82

83 Introduction

84 Herbivorous insects show tremendous variation with regard to dietary specialization. While it is a 85 long-standing assumption that phytochemicals may restrict and direct the evolution of host plant 86 use1, the explicit role of phytochemicals as drivers of host plant associations has been revealed in 87 only a few systems2–5. Proposed mechanisms of how plant secondary compounds could mediate 88 insect-plant interactions include physiological trade-offs in the efficiency of host plant use 89 between generalists and specialists1,6–9. Alternatively, it has been shown that novel host plant 90 associations can create enemy-free spaces for herbivores10,11 either by providing defense11 or 91 refuge from natural enemies10. However, even though it is widely recognized that many insects 92 not only use plants as a dietary resource but also sequester (i.e. absorb and store) plant toxins to 93 defend themselves against predators12–14, the extent to which sequestration could drive the 94 evolution of insect-host plant associations has rarely been addressed4,15. 95 While it has been hypothesized that dietary specialization and sequestration of plant 96 toxins can lead to an evolutionary dead end4,16, there is evidence that ecological specialization 97 does not necessarily prevent host range expansion4. Nevertheless, sequestration and dietary 98 specialization seem to be evolutionarily linked13,17–20, and predators driving the occupation of 99 enemy-free-spaces are typically considered to select for specialization9,21. Recent research 100 indicated that sequestration requires different resistance traits than are required to merely cope 101 with dietary toxins22, suggesting that selection by predators or parasitoids (i.e. the third trophic 102 level) opens a second arena for coevolutionary escalation13,22. Consequently, a rigorous analysis 103 of coevolution between plants and specialized insects requires the integration of adaptations 104 underlying bitrophic interactions with adaptations driven by higher trophic levels13,22,23. 105 Here, we used milkweed bugs (Heteroptera: Lygaeinae) as a model system to test 106 hypotheses about the evolutionary drivers leading to specialized associations with particular plant 107 species. The Lygaeinae comprise about 600 primarily seed-eating species that are well known for 108 their predilection of plants in the Apocynaceae worldwide24–27. Milkweed bugs typically exhibit a 109 red-and-black aposematic coloration and, in addition to defensive scent glands typical for 110 Heteroptera28, several species have been shown to acquire defenses against predators from their 111 host plants29–32. Upon attack, many milkweed bug species release sequestered toxins in a 112 defensive secretion from a specialized storage compartment of the integument (the dorsolateral 113 space)33,34. The large milkweed bug (Oncopeltus fasciatus (Dallas, 1852)) in particular has been 114 studied in detail with regard to sequestration of cardiac glycosides, which it derives from seeds of 115 milkweeds in the Apocynaceae genus Asclepias35. 116 Cardiac glycosides are important defense metabolites of plants in the Apocynaceae, and 117 evolved convergently in at least 11 additional botanical families36. Both compound subtypes, the 118 cardenolides and the bufadienolides, are specific inhibitors of the ubiquitous animal enzyme 119 Na+/K+-ATPase. Specialized insects from at least six taxonomic orders, including several 120 lygaeine species37–40, tolerate cardenolides by expressing Na+/K+-ATPases with several amino 121 acid substitutions that mediate a high degree of cardenolide resistance in vitro (target site 122 insensitivity)40,41. Duplication of the gene coding for the α-subunit of Na+/K+-ATPases in the 123 lygaeine bugs O. fasciatus and Lygaeus kalmii Stal, 1874 resulted in three different copies with 124 up to four amino acid substitutions in regions of the protein critical for cardiac glycoside 125 binding38,39,42, rendering the enzyme resistant to cardiac glycosides42. In addition to resistance, 126 sequestration requires accumulation of toxins from the dietary resource, and milkweed bugs

127 possess an as-of-yet unidentified mechanism for the transport of toxins across the gut epithelium. 128 In summary, milkweed bugs possess a suite of traits related to sequestration and defense that 129 includes aposematic coloration, resistant Na+/K+-ATPases, and mechanisms for accumulation, 130 storage, and release of toxins. This suite of traits may also function as a physiological 131 preadaptation facilitating the sequestration of novel toxin compounds. For example, the milkweed 132 bug Neacoryphus bicrucis (Say, 1825) sequesters pyrrolizidine alkaloids31, a class of compounds 133 unrelated to cardiac glycosides. 134 Sequestration of cardiac glycosides by lygaeine bugs was initially described for O. 135 fasciatus and L. kalmii feeding on Asclepias species43, as well as for Caenocoris nerii (Germar, 136 1847) and Spilostethus pandurus Scopoli, 1763 on oleander (Nerium oleander), both belonging to 137 the Apocynaceae44. A broad survey based on dried museum specimens demonstrated the presence 138 of cardiac glycosides in many genera of Lygaeinae24, suggesting that sequestration of cardiac 139 glycosides is a common trait of milkweed bugs. Furthermore, an evolutionary analysis revealed 140 that sequestration of cardiac glycosides, target site insensitivity of Na+/K+-ATPase, as well as an 141 association with apocynaceous plants are likely ancestral traits of the group40. Nevertheless, 142 despite being specialized to cardiac glycosides, several milkweed bug species are dietary 143 generalists and feed on seeds from a great variety of plant families. The palaearctic species 144 Lygaeus equestris (Linnaeus, 1758) for example, was observed feeding on more than 60 plant 145 species from roughly 20 families45. Even though dietary breadth can vary among species and 146 larval instars to some extent26,45, this large number of potential host plants is in stark contrast to 147 the narrow set of species that produce cardiac glycosides for sequestration. The extreme 148 prevalence of cardiac glycoside sequestration in milkweed bugs thus suggests an important role 149 of selective pressure exerted by higher trophic levels in shaping milkweed bug-host plant 150 associations. 151 Remarkably, several palaearctic species of Lygaeinae are regularly found on plants which 152 are phylogenetically disparate from the Apocynaceae, but which convergently produce cardiac 153 glycosides. Horvathiolus superbus (Pollich, 1779) was observed feeding on Digitalis purpurea 154 (Plantaginaceae) and seems to depend on this plant at least in parts of its distributional range26,46. 155 In addition, we found this species using the cardenolide-producing Erysimum crepidifolium 156 (Brassicaceae) as a host plant in a Digitalis-free habitat. Early instars of L. equestris larvae in 157 Sweden feed exclusively on Vincetoxicum hirundinaria (a cardenolide-free Apocynaceae) and 158 the cardenolide producing Ranunculaceae Adonis vernalis45,47 which is also an important host 159 plant elsewhere26. The generalist milkweed bug S. pandurus was recorded on Urginea maritima 160 (Asparagaceae)48 which produces cardiac glycosides of the bufadienolide-type49. 161 Surprisingly, a closely related species, Spilostethus saxatilis (Scopoli, 1763), which uses a 162 great variety of host plants26,48, is not known to visit cardiac glycoside producing plants except 163 for Asclepias syriaca50 that is not native in its distributional range. However, we and others26,51 164 have often observed this species on flowers and fruits of meadow saffron (Colchicum autumnale, 165 Colchicaceae) which is highly toxic due to the production of alkaloids such as colchicine. 166 Colchicum alkaloids inhibit polymerization of tubulin52, thus showing a mode of action that is 167 different from cardiac glycosides. Nonetheless, based on its evolutionary history, S. saxatilis may 168 be preadapted regarding some traits of the ‘sequestration suite’ including aposematic coloration, 169 storage and release of toxins, and a putative mechanism for transport. In a scenario of 170 sequestration-driven host associations and selection by higher trophic levels, colonization of new 171 sources of potent toxins for sequestration could represent a mode of coevolutionary escalation.

172 Using these insect-plant interactions as a model system, we tested if three species of 173 milkweed bugs sequester cardenolides from four independently colonized, cardiac glycoside- 174 containing plants in the Asparagaceae, Brassicaceae, Plantaginaceae, and Ranunculaceae, and 175 furthermore tested if S. saxatilis sequesters alkaloids from C. autumnale. After confirming the 176 tolerance and sequestration of colchicum alkaloids by S. saxatilis, we evaluated the degree of 177 association between S. saxatilis and C. autumnale by screening 30 S. saxatilis museum specimens 178 from 11 countries for the presence of both colchicum alkaloids and cardenolides. We also 179 assessed if there is an ecological trade-off between sequestration of colchicum alkaloids and 180 cardiac glycosides in this species. Next, we evaluated the importance of cardiac glycoside- or 181 alkaloid-bearing seeds as a dietary resource for milkweed bugs by quantifying larval growth of H. 182 superbus, L. equestris, and S. saxatilis fed with different combinations of toxic- and non-toxic 183 seeds. We then contrasted the effects on growth with a quantification of defensive benefits gained 184 by consumption and sequestration of cardiac glycosides and colchicum alkaloids in H. superbus, 185 L. equestris, S. pandurus, and S. saxatilis against insect (lacewing larvae, Chrysoperla carnea) 186 and avian predators (great tits, Parus major). Finally, we tracked the evolution of colchicine 187 resistance in a phylogenetic framework using cardenolide and colchicine injection assays to 188 mimic sequestration. 189 We suggest that sequestration of plant toxins as a defense against predators mediates 190 specialized associations of milkweed bugs with specific host plant species. The preadaptations for 191 sequestration of cardiac glycosides and their convergent occurrence within distantly related plants 192 most likely facilitated new host associations. Furthermore, a subset of these preadaptations may 193 have facilitated the shift to new host plants with functionally distinct but highly potent toxins. We 194 thus demonstrate that species which are dietary generalists under a bitrophic perspective may 195 nonetheless be highly specialized on plants that provide defenses against the third trophic level. 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 5

218 Material and Methods

219 Field-sampling of milkweed bugs for chemical analysis 220 In order to assess sequestration of toxins under natural conditions, adult milkweed bugs were 221 collected in the field from habitats with natural stands of their toxic host plants (Figure 1, 222 Supplemental Figure 6). Field work in protected areas was permitted by the responsible agencies 223 (see acknowledgments). Notes on natural history observations are given in the electronic 224 supplementary material. S. saxatilis were collected in Nüstenbach (August 15th, 2015) and 225 Berghausen (September 18th, 2015), Baden-Württemberg, Germany, either from C. autumnale 226 (Colchicaceae) or feeding on other plants. H. superbus was collected on August 15th and 16th, 227 2015 in a Digitalis purpurea (Plantaginaceae) habitat close to Eberbach, Baden-Württemberg, 228 Germany. We collected the same species feeding on Erysimum crepidifolium (Brassicaceae) in a 229 Digitalis-free habitat (Schloßböckelheim, Rheinland-Pfalz, Germany) on June 12th and 13th 2018. 230 L. equestris was collected in the nature reserve ‘Oderhänge Mallnow’ north of Lebus, 231 Brandenburg, Germany on April 19th, 2016 during the blooming of Adonis vernalis 232 (Ranunculaceae). Lastly, S. pandurus was collected from infructescences of Urginea maritima 233 (Asparagaceae) in ‘Parque Natural Sierra de Aracena y Picos de Aroche’, Aracena, Spain, in 234 early November 2016. After bringing the bugs to the lab, they were maintained on sunflower 235 seeds and water provided in Eppendorf tubes plugged with cotton under ambient conditions to 236 purge their guts from remaining toxins. After 14 days (≥ 12 days for S. pandurus), bugs were 237 frozen at -80°C and freeze-dried for chemical analysis as described below. 238 239 Origin and maintenance of bugs used for experiments 240 Specimens of L. equestris, H. superbus, and S. pandurus used for seed mixture feeding, 241 predation, and injection experiments were obtained from laboratory colonies maintained for 242 several generations at the time of the experiments. Specimens for founding colonies were 243 collected at the locations described above (H. superbus from Eberbach), except for S. pandurus 244 used in injection and predation assays that were from a lab-strain originating from Portugal 245 collected in 2016. Oncopeltus fasciatus used as controls for injection assays were from a long- 246 term laboratory colony (origin United States), obtained from the University of Hamburg in 2015. 247 Bugs were raised on husked sunflower seeds and supplied with water (see above) as well as 248 pieces of cotton for oviposition. Colonies were maintained in environmental chambers at 28°C 249 and 60% humidity at a light: dark cycle of 16 h: 8 h. Specimens of S. saxatilis were raised from 250 eggs obtained from adults collected in the field (Berghausen) that were fed with sunflower seeds 251 under ambient conditions. Pyrrhocoris apterus (Linnaeus, 1758) used as a non-adapted outgroup 252 for injection assays were collected in the field (Giessen, Germany) and used directly or after one 253 day of maintenance in the laboratory on linden seeds. 254 255 Seed mixture experiments 256 To assess if inclusion of Digitalis-, Adonis-, or Colchicum-seeds (‘toxic seeds’ hereafter) in the 257 dietary spectra improves growth, we reared larvae of S. saxatilis, L. equestris, and H. superbus 258 under four dietary treatments. Starting with pre-weighed first instar larvae, bugs were raised 259 either on sunflower seeds only (positive control), a seed mixture comprising 11-15 natural host 260 plant species to reflect the broad dietary spectra of S. saxatilis and L. equestris (Supplemental 261 Table 1), the identical seed mixture supplemented with toxic seeds, or toxic seeds only. To

262 establish species-specific seed mixtures, we selected natural host plant species based on literature 263 data or own field observations. Toxic seeds were selected based on natural associations of the 264 bug species with toxic plants in the field. Specifically, we used seeds of D. purpurea for H. 265 superbus, seeds of A. vernalis for L. equestris, and seeds of C. autumnale for S. saxatilis. 266 Untreated, ripe seeds were either obtained commercially or collected in the field (Digitalis 267 purpurea, Eberbach 2016). Seed mixtures consisted of 11 host species from 4 botanical families 268 for S. saxatilis and 15 host species from 7 botanical families for L. equestris. Due to the lack of 269 natural history data for H. superbus, we used the seed mixture used for S. saxatilis also for H. 270 superbus. Within seed mixtures, we standardized proportions of individual plant species based on 271 mass. Growth experiments were carried out in spatially randomized Petri dishes in a growth 272 chamber (Binder KBWF 240, Tuttlingen, Germany) under the following conditions: 16: 8 h 273 day/night cycle, 26 °C (S. saxatilis) or 28 °C (L. equestris and H. superbus), and 60 % humidity 274 over the course of three weeks. We lined Petri dishes (60 mm x 15 mm, with vents, Greiner Bio- 275 One, n = 11 for all species and diets) with filter paper and added three 1st instar larvae to each 276 dish. Petri dishes were supplied with a water source (see above) and 140.7 mg (± 0.72 SE, S. 277 saxatilis), 145.1 mg (± 1.03 SE, L. equestris), or 140.1 mg (± 0.69 SE, H. superbus) seeds. 278 We recorded body mass weekly by anesthetizing all bugs of a Petri dish with CO2 and 279 weighing them jointly. In addition, we recorded survival of bugs weekly. After the experiment, 280 we transferred at least one bug from each Petri dish to fresh sunflower seeds for a period of two 281 weeks to clean guts from potentially remaining toxins. Note that not all bugs had reached the 282 adult stage at the time of transfer, thus the time spent on sunflower seeds during the adult stage 283 varied and some bugs were still larvae at the time of analysis (see Supplemental Table 2 and the 284 supplementary results). Finally, bugs were frozen at -80°C, freeze dried, weighed, and analyzed 285 to quantify sequestered toxins via high performance liquid chromatography (HPLC) as described 286 below (n = 11 for all species and diets, except of H. superbus on the seed mixture and pure 287 Digitalis seeds and L. equestris on the seed mixture containing Adonis seeds, n = 10). 288 Growth of bugs on different seed mixtures, approximated as body mass (i.e. the total 289 weight of all bugs per Petri dish divided by the number of remaining individuals) after three 290 weeks of feeding, was analyzed by ANCOVA using the standard least squares method in JMP 291 (JMP v. 13, SAS Institute, Cary, NC, USA). Samples sizes were 11 (i.e. 11 Petri dishes with up to 292 three bugs) for all species and treatments except of H. superbus on the seed mixture and pure D. 293 purpurea seeds (n = 10). Body masses were log10-transformed to achieve homogeneity of 294 variances and normality of residuals. Dietary treatment was treated as a main effect, and initial 295 mass of the bugs was included as a covariate. Two individuals of L. equestris (Adonis treatment) 296 and of H. superbus (Digitalis and seed mixture plus Digitalis treatment) were statistical outliers, 297 but their exclusion did not affect the direction or significance of effects. In addition to comparing 298 final body mass after three weeks of feeding, we also modelled growth as a continuous process 299 (see supplementary electronic material). We analyzed developmental time across treatments by 300 comparing the number of days (log10-transformed) after which at least one bug per petri dish 301 reached the adult stage using ANOVA in JMP (n = 11 for L. equestris on all treatments; n = 9 for 302 H. superbus on the seed mixture without Digitalis-seeds, and n = 10 for all other treatments; note 303 that some individuals turned into adults after being transferred to sunflower seeds). We omitted 304 this analysis for S. saxatilis since on the Colchicum diet only one individual turned adult during 305 the time of observation. 306

307 Preparation of insect specimens for HPLC analysis 308 We determined the concentration of plant toxins sequestered by individual milkweed bugs with 309 HPLC-DAD. We added 1 ml methanol (HPLC grade) containing 0.01 mg of the internal standard 310 digitoxin (Sigma-Aldrich, Taufkirchen, Germany) to freeze-dried specimens of L. equestris and 311 S. pandurus. For H. superbus, digitoxin was replaced by oleandrin (Phytolab, Vestenbergsgreuth, 312 Germany) due to the natural occurrence of digitoxin in Digitalis. For specimens of S. saxatilis 313 and S. pandurus, we used no internal standard and quantified colchicum alkaloids and 314 bufadienolides with an external calibration curve (see below). After the addition of ca. 900 mg 315 zirconia beads (Roth, Germany), specimens were homogenized in a Fast-Prep-24 homogenizer 316 (MP Biomedicals, Germany) for two 45 sec cycles at a speed of 6.5 m/sec. After centrifugation at 317 16,100 x g for 3 min, supernatants were transferred to fresh 2 ml plastic vials (Sarstedt, 318 Germany). Extractions were repeated once more with pure methanol. Pooled supernatants were 319 evaporated under nitrogen gas. Subsequently, we re-suspended samples by adding 100 µl of 320 methanol (200 µl for field-collected S. pandurus and saxatilis) and agitation in the Fast-Prep-24 321 homogenizer (45 s, 6.5 m/sec) without beads to facilitate dissolution of dried residues. Finally, 322 samples were centrifuged (16,100 x g, 3 min) and filtered into HPLC-vials using Rotilabo®- 323 syringe filters (nylon, 0.45 µm, Roth, Germany). Eggs obtained from S. saxatilis (n = 5; pools of 324 7, 18, 22, 4, and 7 eggs) females collected in Berghausen on May 5th 2016 and from field- 325 collected L. equestris (Lebus, n = 3; pools of 27, 28, and 63 eggs) were freeze dried and extracted 326 as described above with 2 x 500 µl or 2 x 1 ml methanol, respectively. Details on harvesting of 327 haemolymph and defensive secretion (i.e. clear droplets released at the integument upon attack) 328 of S. saxatilis, as well as the preparation of dried museum specimens and plant seeds for chemical 329 analysis are described in the electronic supplementary material.

330 HPLC analysis of cardenolides, bufadienolides, and colchicum alkaloids 331 Fifteen microliters of extract were injected into an Agilent 1100 series HPLC and compounds ® 332 were separated on an EC 150/4.6 NUCLEODUR C18 Gravity column (3 µm, 150 mm x 4.6 mm, 333 Macherey-Nagel, Düren, Germany). Cardenolides and bufadienolides were eluted at a constant 334 flow of 0.7 ml/min at 30°C with an acetonitrile–H2O gradient as follows: 0–2 min 16% 335 acetonitrile, 25 min 70% acetonitrile, 30 min 95% acetonitrile, 35 min 95% acetonitrile, 37 min 336 16% acetonitrile, reconditioning for 10 min at 16% acetonitrile. We recorded UV absorbance 337 spectra from 200 to 400 nm with a diode array detector. Peaks with symmetrical absorption 338 maxima between 216 and 222 nm were interpreted as cardenolides, integrated at 218 nm and 339 quantified based on the peak area of the known concentration of the internal standards digitoxin 340 or oleandrin. Peaks with a symmetrical absorption maximum of 300 nm were interpreted as 341 bufadienolides. For bufadienolide analysis, we used an external calibration curve based on 342 proscillaridin A. 343 Colchicum alkaloids were eluted at a constant flow of 0.7 ml/min at 30°C with an 344 acetonitrile–0.25 % phosphoric acid gradient as follows: 0–2 min 10% acetonitrile, 10 min 40% 345 acetonitrile, 15 min 80% acetonitrile, 16 min 10% acetonitrile, reconditioning for 5 min at 10% 346 acetonitrile. UV absorbance spectra were recorded from 190 to 400 nm by a diode array detector. 347 Peaks with absorption maxima at 245 and 350 nm resembling the absorption spectra of colchicine 348 were recorded as colchicosides and quantified at 350 nm. Colchicine-equivalents were calculated 349 based on an external colchicine calibration curve. 350 Analysis of chromatograms was carried out with the Agilent ChemStation software 351 (B.04.03). Details on the evaluation of individual datasets are described in the electronic 8

352 supplementary material. Individual compounds were identified by comparisons of UV spectra 353 and retention time with commercial reference compounds and liquid chromatography – mass 354 spectrometry (details are described in the electronic supplementary material). 355 For analyzing sequestered toxins during seed feeding assays we used Welch’s test (JMP) 356 and the Game-Howell post-hoc test (see www.biostathandbook.com) on log10 transformed data to 357 evaluate differences across treatments, since data for L. equestris and S. saxatilis did not meet the 358 assumption of equal variance. For this analysis, we excluded all data for individuals of H. 359 superbus and L. equestris raised on sunflower seeds or seed mixtures without toxic seeds, as 360 these lacked sequestered cardenolides.

361 Behavioral assays with lacewings 362 We used larvae of the lacewing Chrysoperla carnea (Neuroptera, Chrysopidae) to test the effect 363 of sequestered plant metabolites against an arthropod predator. Eggs of L. equestris, H. superbus, 364 and S. pandurus were transferred to either sunflower seeds (non-toxic controls) or toxic seeds (A. 365 vernalis for L. equestris, D. purpurea for H. superbus, and U. maritima for S. pandurus). Petri 366 dishes were supplied with a water source as described above. After at least 3 (S. pandurus) to 5 367 (H. superbus) days in a growth chamber (Binder KBWF 240) at a 16: 8 h day/night cycle, 28 °C, 368 60% humidity, 1st to 3rd instar larvae (mainly 1st and 2nd) were presented individually to a 369 lacewing larva (2nd or 3rd instar). Larvae hatched from eggs of field-collected S. saxatilis were 370 maintained on C. autumnale seeds for at least 4 days as described above. Due to the maternal 371 transfer of colchicum alkaloids into the eggs, we used O. fasciatus larvae raised on sunflower 372 seeds as a non-toxic negative control for these assays. We saved at least five individuals of each 373 species from both treatments (toxic seeds and non-toxic sunflower seeds) to assess the amount of 374 toxins sequestered using HPLC-DAD as described above. 375 Lacewing larvae were obtained commercially (Sautter & Stepper GmbH, Ammerbuch, 376 Germany), maintained individually on Sitotroga (Lepidoptera: Gelechiidae) eggs (Katz Biotech 377 AG, Baruth, Germany) for 3-5 days at room temperature and starved for two days before the 378 experiments. In a first set of experiments, one milkweed bug larva was exposed to one lacewing 379 larva in a Petri Dish (60 mm x 15 mm, with vents) and observed until the lacewing larvae 380 attacked for the first time. After the first attack was over, lacewing larvae were removed and 381 milkweed bug larvae were provided with a sunflower seed and water and checked for survival the 382 next day. We excluded trials in which lacewing larvae did not attack bug larvae over the time of 383 observation. The proportion of surviving milkweed bug larvae after the first attack by a lacewing 384 across treatments was compared using two-tailed Fisher’s exact test in JMP. 385 In a second set of experiments, we assessed the effect of sequestered toxins on 386 consumption of milkweed bug larvae by lacewing larvae. For L. equestris, both experiments 387 (survival and feeding) were carried out twice since we initially assumed that the lack of an effect 388 was due to the thick wall of the Adonis vernalis follicle rendering the seed inaccessible to small 389 L. equestris larvae. Therefore, we repeated the experiments and maintained larvae on A. vernalis 390 follicles chopped with a razor blade. Due to an overall lack of effect on lacewing behavior (see 391 below), both experiments were combined for analyses. Details of this experiment are described in 392 the electronic supplementary material.

393 Behavioral assays with birds 394 Adults of S. saxatilis (specimens from Berghausen, see above), L. equestris and H. superbus 395 (specimens from Eberbach, see above), either raised on sunflower seeds (control) or on toxic 9

396 seeds (C. autumnale for S. saxatilis, A. vernalis for L. equestris, and D. purpurea for H. 397 superbus), were offered to hand-reared juvenile great tits (Parus major; Passeriformes: Paridae) 398 to test whether sequestered toxins protect bugs against avian predators. For this purpose, we 399 raised first and second instar larvae to adults either on pure sunflower seeds or on a 1:1 mixture 400 of sunflower with toxic seeds in plastic containers and supplied them with water as described 401 above. Larvae were maintained in a growth chamber (Fitotron® SGC 120, Weiss Technik, 402 Loughborough, UK) at a 16: 8 h day/night cycle, at 26 - 27°C, and 60 % humidity. Before the 403 experiment, we transferred bugs to pure sunflower seeds and maintained them at least for one 404 week under the same conditions as described above to purge their guts from potentially retained 405 plant toxins. 406 Birds were tested individually in wood-frame cages (70 × 70 × 70cm) with wire-mesh walls, 407 equipped with a perch, a water bowl and a feeding tray. Cages were illuminated by daylight- 408 simulating Osram Biolux 18-W/965 tubes. Prey was offered to birds in glass Petri dishes 409 (diameter 50 mm) placed on the feeding tray. This way, all prey items appeared conspicuous 410 against the light-colored background of the wooden tray. Before the experiment, the birds were 411 habituated to the experimental cage, trained to eat mealworms from the tray, and deprived of food 412 for 2 hours. We observed the birds through a one-way glass in the front wall of the cage and 413 recorded their behavior using the Observer XT (Noldus) software. Details of hand-rearing 414 juvenile birds are described in the electronic supplementary material. 415 Birds (100 in total) were divided into 3 experimental groups: 40 birds were tested with S. 416 saxatilis, 40 birds with L. equestris and 20 birds with H. superbus. Within each group, half of the 417 birds were tested with bugs raised on seeds of their respective toxic host plants and the other half 418 with bugs raised on sunflower seeds (non-toxic control). To account for the effects of prey 419 novelty on bird responses we used larvae of Jamaican field crickets (Gryllus assimilis) of a 420 similar size as the bugs as a palatable prey control that would be unfamiliar to the birds. The 421 experiment consisted of six five-minute trials (following immediately one after another) in which 422 the birds were alternately offered a milkweed bug or a cricket, starting with the cricket. In each 423 trial we recorded if the prey was attacked (pecked or seized), killed and eaten (at least partly), 424 latency of the first attack, duration of prey handling, and number and duration of discomfort- 425 indicating behavior (beak wiping and head shaking; supplementary results). If the bug was 426 attacked but alive at the end of the trial, it was provided with water and sunflower seeds and 427 checked for survival on the next day. 428 Bird predation data were analyzed using generalized linear models in R53. Attack rates 429 and survival rates of the bugs were compared across the three trials using generalized estimating 430 equation models (GEE, package geepack54) with binomial errors. Trial number, bug species and 431 host plant toxicity were entered as fixed effects and bird individual (id) as a random effect. The 432 models initially included all possible two-way interactions and were simplified by comparing 433 nested models using the Quasi-information criterion (QIC). In the analysis of survival rates, only 434 the data from bugs attacked in respective trials were included. To find out whether the general 435 effect of host plant toxicity on reactions of birds also holds for each of the milkweed bug species 436 studied, the abovementioned models were run separately for each bug species. Besides attack and 437 survival rates, we also analyzed attack latencies, durations of discomfort-indicating behavior of 438 birds, whether the bugs were (at least partly) eaten, and survival of bugs compared to control 439 crickets; see the electronic supplementary material for details.

440 10

441 Injection experiments to assess cardenolide and colchicine resistance 442 We injected adults of field-collected Pyrrhocoris apterus, or sunflower-raised O. fasciatus, S. 443 saxatilis (Berghausen), and S. pandurus (Portugal) with colchicine or the cardenolide ouabain to 444 test for the ability to tolerate these toxins in the body cavity (i.e. to mimic sequestration). We 445 injected one microliter of toxins (dissolved in phosphate buffered saline, PBS, pH 7.4) or PBS as 446 a control with glass capillary needles. Solutions were injected laterally between the penultimate 447 and the last abdominal segment using a micromanipulator and a microsyringe pump injector 448 (World Precision Instruments, Sarasota, FL, USA) under a dissecting microscope (n = 8 449 individuals per dose). 450 In total, we carried out three injection experiments. In a first experiment, P. apterus, O. 451 fasciatus, and S. saxatilis were injected with either a high dose of ouabain (5 mg/ml) or 452 colchicine (10 mg/ml, Figure 4). As we observed no effect of colchicine in S. saxatilis as opposed 453 to the other species, we injected S. saxatilis with an even higher dose of colchicine in a second 454 experiment and again injected 5 mg/ml ouabain in additional specimens for comparison. Since P. 455 apterus responded to both toxins in the first trial, we injected 0.1, 1, 5, or 10 mg/ml colchicine or 456 0.1, 1, 2.5, 5 mg/ml ouabain (i.e. 0.1 to 10 µg/individual and 0.1 to 5 µg/individual, respectively) 457 within the same attempt to address the extent of resistance quantitatively. O. fasciatus that 458 tolerated 5 mg/ml ouabain were only injected with increasing concentrations of colchicine (0.1, 1, 459 5 and 10 mg/ml, i.e. 0.1 to 10 µg/individual). To keep the number of injected animals as low as 460 possible we omitted injections of PBS during experiment 2 (tolerance to injections per se was 461 already apparent from the first experiment). Lastly (experiment 3), we injected S. pandurus, a 462 congener of S. saxatilis with PBS, 0.1, 1, 5, or 10 mg/ml colchicine. After injection, we 463 maintained bugs individually in Petri dishes with a water source (see above) and one sunflower 464 seed under ambient conditions. On the next day, we assessed bugs for signs of paralysis (i.e. 465 inability to walk, slowed movement of legs and antennae). Injection assays with one dose of a 466 toxin were analyzed using Fisher’s exact test in JMP with Bonferroni correction for multiple 467 comparisons. For comparing dose-dependent effects, we used the Cochrane-Armitage trend test 468 in JMP.

469 Statistical analyses 470 For better readability, statistical analyses are explained at the end of each methods section (see 471 above). If not mentioned in the main text, sample sizes are reported in the figure legends.

472 Results

473 Growth of milkweed bugs on different diets 474 We tested if the availability of ‘toxic seeds’ influences larval development. After three weeks, a 475 diet of pure sunflower seed resulted in maximal growth in all three insect species tested (Figure 476 2a-c). We found a significant effect of seed diet on growth of L. equestris (F3,39 = 56.431, p < 477 0.001) and S. saxatilis (F3,39 = 56.631, p < 0.001), but not of H. superbus (F3,37 = 0.425, p = 478 0.736) which grew equally across all treatments. On diverse seed mixtures without toxic seeds, S. 479 saxatilis grew as well as on sunflower seeds (LSMeans Tukey HSD: p = 0.883) while L. equestris 480 only reached about half of the body mass compared to the sunflower diet (LSMeans Tukey HSD: 481 p < 0.001). Inclusion of ‘toxic seeds’ into these mixtures did not increase growth for either of the 482 species (LSMeans Tukey HSD: L. equestris, p = 0.384; S. saxatilis, p = 0.936). In fact, diets 483 consisting exclusively of toxic seeds resulted in lower body mass compared to seed mixtures for 11

484 both species (LSMeans Tukey HSD: L. equestris, p < 0.001; S. saxatilis, p < 0.001), and in a 485 reduction of > 50 % in body mass compared to the sunflower diet (LSMeans Tukey HSD: L. 486 equestris, p < 0.001; S. saxatilis, p < 0.001). Initial body mass affected the final weight of S. 487 saxatilis (F1,39 = 5.453, p = 0.025) but not of L. equestris (F1,39 = 0.034, p = 0.855) and H. 488 superbus (F1,37 = 0.051, p = 0.823). The modelling of larval growth as a continuous process and 489 comparison of absolute growth rates revealed similar results (Supplemental Figure 1. We found 490 negative effects of Adonis seeds on developmental time in L. equestris but not of Digitalis seeds 491 in H. superbus (supplementary results).

492 Sequestration in milkweed bugs on different diets 493 In addition to growth, we also quantified sequestration of cardenolides and colchicum alkaloids. 494 We found substantial sequestration of cardenolides in H. superbus and in L. equestris raised on 495 pure Digitalis or pure Adonis seeds and seed mixtures containing toxic seeds, respectively, while 496 bugs raised on sunflower seeds or seed mixtures without toxic seeds were devoid of cardenolides 497 (Figure 2d,e). Similarly, S. saxatilis raised on seeds of Colchicum and seed mixtures containing 498 Colchicum seeds sequestered high amounts of colchicum alkaloids (Figure 2f). Bugs from diet 499 treatments lacking Colchicum seeds also contained low levels of these toxins which are most 500 likely derived from field-collected females by transfer via the egg. The amount of sequestered 501 toxins was always highest on the diets comprised of toxic seeds only (Welch’s test of diet effect; 502 H. superbus: F1,19 = 18.863, p < 0.001; L. equestris: F1,10 = 66.206, p < 0.001; S. saxatilis: F3,20 = 503 83.568, p < 0.001).

504 Sequestration of plant toxins in field-collected milkweed bugs 505 We collected adult milkweed bugs in the field to test for sequestration of plant toxins under 506 natural conditions (Supplemental Figure 6). All specimens of H. superbus (n = 10) from a habitat 507 with D. purpurea contained cardenolides (Figure 1c) ranging from 1 to 8.9 µg per mg dry mass 508 (2.6 to 28.6 µg per insect, n = 10). H. superbus collected from a Digitalis-free habitat (n = 12) 509 that we observed feeding on pods of E. crepidifolium invariably contained high amounts of 510 sequestered cardenolides (Figure 1c) ranging from 23 – 61.2 µg/mg (41 – 147 µg per specimen). 511 All L. equestris obtained from the A. vernalis site had cardenolide concentrations ranging from 512 0.14 to 28.12 µg/mg dry mass (4.7 to 459.6 µg per individual, n = 12; Figure 1b). Moreover, we 513 detected cardenolides in eggs laid by field-collected females (0.23 µg/mg dry weight ± 0.01, n = 514 3; mean ± SE). S. pandurus collected from infructescences of U. maritima contained up to 17.8 515 µg bufadienolides per mg dry mass (up to 808 µg per individual, n = 7; Figure 1b). 516 Adults of S. saxatilis obtained from two different populations consistently contained 517 colchicum alkaloids (Figure 1a) ranging from 0.05 – 6.2 µg/mg dry mass (1.55 – 113.7 µg per 518 individual, Nüstenbach, n = 16) and 6.4 – 9.4 µg/mg dry mass (134.3 – 214.3 µg per individual, 519 Berghausen, n = 9). Eggs from field-collected females contained colchicum alkaloids (0.94 520 µg/mg dry weight ± 0.26, n = 5; mean ± SE). Concentrations of colchicum alkaloids in S. 521 saxatilis defensive secretion were more than 50 times higher compared to haemolymph (paired t- 522 test on six paired samples: t = 4.234, df = 5, p = 0.008, Supplemental Figure 4). The comparison 523 of sequestered toxins to toxins present in the host plant seeds revealed a clear overlap of 524 individual compounds only in some insect species (Supplemental Figure 3) indicating extensive 525 metabolism or selective uptake and up-concentration by the other species (see the electronic 526 supplementary material for details on structural identification and comparison to seed extracts).

527 Colchicum alkaloids in museum specimens 528 We screened 30 museum specimens of S. saxatilis from 21 locations in 10 European countries 529 and one location in North Africa (Supplemental Figure 5). Although some of the specimens were 530 more than 110 years old, we detected substantial amounts of colchicum alkaloids ranging from 531 0.8 µg to 182.5 µg per individual (58 ± 8.69, mean ± SE) in all specimens. Remarkably, only two 532 of the specimens tested contained trace amounts of putative cardenolides, suggesting that 533 sequestration of cardenolides does not play a role for the defense of S. saxatilis.

534 Effect of sequestered toxins on lacewing predation 535 We assessed the effect of sequestered plant toxins on lacewing predation in four species of 536 milkweed bugs. H. superbus larvae raised on seeds of Digitalis contained 3.15 ± 0.99 µg 537 cardenolides per mg dry weight (mean ± SE, n = 7) and survived lacewing attacks more often 538 compared to sunflower-raised individuals (Figure 3a; p = 0.02, two tailed Fisher’s exact test). 539 However, even though L. equestris larvae raised on Adonis seeds contained higher cardenolide 540 amounts (whole seeds: 4.95 ± 0.27 µg/mg, n = 8; chopped seeds: 5.01 ± 0.45 µg/mg, n = 5; mean 541 ± SE), only one sunflower-raised larva out of a total of 63 L. equestris larvae tested (n = 32 for 542 sunflower, n = 31 for Adonis seeds) survived the first attack by a lacewing larva (Figure 3b; p = 543 1, two tailed Fisher’s exact test). Similarly, S. pandurus larvae raised on Urginea seeds contained 544 16.11 ± 4.39 µg/mg bufadienolides (n = 7), yet only one Urginea-raised larva out of all 40 tested 545 survived the first lacewing attack (n = 20 for both treatments, p = 1, two tailed Fisher’s exact 546 test). In contrast, sequestration of colchicum alkaloids protected S. saxatilis from lacewing 547 attacks. Thirteen out of 20 larvae derived from eggs of field-collected females survived the first 548 attack by a lacewing larva (Figure 3c) while there was no survival in sunflower raised Oncopeltus 549 larvae used as a control (p < 0.001, two tailed Fisher’s exact test). Concentrations of colchicum 550 alkaloids in S. saxatilis larvae were 53.15 ± 3.12 µg/mg (n = 10, mean ± SE). Even though only 551 two out of four systems tested showed an effect on survival, there was at least some evidence for 552 a negative effect of sequestered compounds on consumption and weight gain by lacewing larvae 553 in all four systems (Supplemental Figure 7, see the electronic supplementary material for details). 554 555 Effects of sequestration on defence against avian predators 556 We analysed the effects of sequestration of host plant toxins on defence against avian predators in 557 L. equestris, S. saxatilis and H. superbus. Overall attack rates were similar for all bug species 2 558 (GEE, χ 2 = 3.242, P = 0.198), but significantly lower when the birds were tested with bugs from 2 559 toxic host plants (GEE, χ 1 = 5.596, P = 0.018). Attack rates decreased over the three successive 2 560 trials (GEE, χ 1 = 115.785, P ˂ 0.001), and the decrease was affected neither by bug species 2 561 (GEE, trial: species interaction, χ 2 = 2.032, P = 0.362) nor host plant toxicity (GEE, trial: host 2 562 plant interaction, χ 1 = 0.815, P = 0.367). However, a difference in this decrease of attack rates 563 became apparent if species were analysed separately. In S. saxatilis, attack rates decreased over 2 564 trials (GEE, χ 1 = 55.968, P ˂ 0.001), and more so when the bugs were coming from Colchicum 2 565 autumnale than from sunflower (GEE, hostplant, χ 1 = 2.414, P = 0.121; trial: host plant 2 566 interaction, χ 1 = 6.009, P = 0.014). Likewise, attack rates towards H. superbus decreased over 2 567 trials (GEE, χ 1 = 39.327, P ˂ 0.001), and the decrease was steeper for bugs raised on Digitalis 2 2 568 purpurea (GEE, host plant, χ 1 = 1.250, P = 0.412; trial: host plant interaction, χ 1 = 6.525, P = 569 0.011). Contrastingly, attack rates towards L. equestris decreased over trials irrespectively to host 2 2 570 plant toxicity (GEE, trial, χ 1 = 28.957, P ˂ 0.001; hostplant, χ 1 = 0.056, P = 0.813; trial: host 2 571 plant interaction, χ 1 = 1.067, P = 0.302). 13

572 Overall survival rates following attack were higher in bugs from toxic host plants (GEE, 2 2 573 χ 1 = 24.477, P ˂ 0.001) and increased significantly over successive trials (GEE, χ 1 = 13.581, P 574 ˂ 0.001), with this increase being steeper in the bugs from toxic host plants (GEE, trial: host plant 2 575 interaction, χ 1 = 9.341, P = 0.002; Figure 4d-f). Bug species affected neither the overall survival 2 2 576 (GEE, χ 2 = 0.777, P = 0.678) nor its increase over the trials (GEE, trial: species interaction, χ 1 = 577 3.403, P = 0.183). Separate analyses for each species revealed that survival rate was generally 2 578 higher if milkweed bugs were raised on toxic host plants (GEE, S. saxatilis: χ 1 = 12.471, P ˂ 2 2 579 0.001; L. equestris: χ 1 = 10.037, P = 0.002; H. superbus: χ 1 = 5.086, P = 0.024), and it increased 2 2 580 over the trials (GEE, S. saxatilis: χ 1 = 10.911, P ˂ 0.001; L. equestris: χ 1 = 5.644, P = 0.018; H. 2 581 superbus: χ 1 = 17.788, P ˂ 0.001; Figure 4d-f). The increase was steeper in S. saxatilis and L. 2 582 equestris coming from toxic host plants (GEE, trial: host plant interaction, S. saxatilis: χ 1 = 2 583 6.097, P = 0.014; L. equestris: χ 1 = 4.394, P = 0.036), but there was no significant difference in 2 584 H. superbus (GEE, χ 1 = 2.442, P = 0.118). 585 In addition, sequestered plant toxins increased bird attack latencies across trials 586 (Supplemental Figure 8), the duration of discomfort-indicating behavior (Supplemental Figure 9), 587 and decreased the chance that the birds would consume the bugs. Nevertheless, compared to 588 crickets, milkweed bugs devoid of sequestered toxins were not entirely undefended against the 589 birds, emphasizing the role of endogenous scent-gland secretion (see the electronic 590 supplementary material for details). 591 592 In vivo tolerance to injected ouabain and colchicine 593 Due to the rare occurrence of colchicine in nature which is restricted to Colchicum spp. and other 594 Colchicaceae55, we predicted that S. saxatilis was not preadapted to this toxin but evolved novel 595 resistance traits against colchicine. In addition, we predicted this species to retain resistance 596 against cardiac glycosides based on its evolutionary history. To test for toxin resistance and to 597 mimic sequestration, we injected colchicine or ouabain directly into the body cavity of milkweed 598 bugs. Besides S. saxatilis, we used Pyrrhocoris apterus (Pyrrhocoridae) as a non-adapted 599 outgroup, and the lygaeine Oncopeltus fasciatus as a cardenolide-sequestering milkweed 600 specialist. None of the species tested was affected by blank injections of the solvent PBS (Figure 601 4, Supplemental Figure 10). As expected, P. apterus was unable to tolerate injections of either 5 602 µg ouabain or of 10 µg colchicine (ouabain vs. PBS, p = 0.001; colchicine vs. PBS, p = 0.001; 603 two tailed Fisher’s exact test at p = 0.025 after Bonferroni correction). All O. fasciatus 604 individuals tolerated an injection of 5 µg ouabain but were not able to tolerate 10 µg colchicine (p 605 = 0.001, two tailed Fisher’s exact test). As predicted, S. saxatilis tolerated injections with both 606 classes of toxins (ouabain vs. PBS, p = 1; colchicine vs. PBS, p = 1; two tailed Fisher’s exact test 607 at p = 0.025 after Bonferroni correction). We furthermore found that S. pandurus, a congener of 608 S. saxatilis, was not able to tolerate colchicine injections. Moreover, O. fasciatus responded to 609 colchicine and P. apterus to ouabain and colchicine, in a dose-dependent manner (Supplemental 610 Figure 10). S. saxatilis tolerated injections of up to 30 µg colchicine per animal, the highest dose 611 tested (see the electronic supplementary materials for details). 612

613 614 Figure 1. Association of milkweed bugs with toxic plants and sequestration of toxic plant compounds in field- 615 collected specimens. Left and center: phylogenetic relationships of milkweed bugs40 and host plants of focal species. 616 The red arrow indicates the association of S. saxatilis with the colchicum alkaloid producing C. autumnale 617 (Colchicaceae), blue arrows indicate associations with plants containing cardiac glycosides. From top to bottom: S. 618 saxatilis on C. autumnale, S. pandurus on U. maritima (Asparagaceae), L. equestris on A. vernalis (Ranunculaceae), 619 H. superbus on D. purpurea (Plantaginaceae), and H. superbus on E. crepidifolium (Brassicaceae). Right: 620 Concentrations of plant toxins sequestered by milkweed bugs collected in the field. (a): Colchicum alkaloids in S. 621 saxatilis from two different populations (left: Nüstenbach, n = 11, right: Berghausen, n = 9). (b): Cardenolides in L. 622 equestris from A. vernalis (left, n = 12) and bufadienolides in S. pandurus from U. maritima (right, n = 7). Scale for 623 cardenolides is identical with scale for bufadienolides. (c): Cardenolides in H. superbus from D. purpurea (left, n = 624 10) and in H. superbus from E. crepidifolium (n = 12). Diamonds represent means ± SE, circles represent jittered raw 625 data.

626

627 Figure 2. Growth of milkweed bugs and sequestration of plant toxins across different diets. Larvae of H. superbus 628 (a), L. equestris (b), and S. saxatilis (c) were raised on sunflower seeds, a seed mixture, the same mixture containing 629 either Digitalis- (H. superbus), Adonis- (L. equestris), or Colchicum- (S. saxatilis) seeds, or pure Digitalis-, Adonis-, 630 or Colchicum-seeds. Larval mass was recorded over a period of three weeks. Bars represent means of body mass 631 after three weeks ± SE. Diamonds represent retransformed model means of data that were log10 transformed for 632 statistical analysis. Sample sizes for the obtained body masses for H. superbus were: sunflower = 11, mixture = 10, 633 mixture plus Digitalis = 11, Digitalis = 10; for L. equestris and S. saxatilis: n = 11 for all diets. After the growth 634 experiment, bugs were harvested for chemical analyses. The amount of sequestered cardiac glycosides from Digitalis 635 (d) and Adonis (e) or colchicum alkaloids (f) was always highest on pure diets but sequestration was also substantial 636 in seed mixtures containing Digitalis-, Adonis-, or Colchicum-seeds. Colchicum alkaloids found in bugs raised on 637 sunflower seeds or seed mixtures lacking Colchicum-seeds originate from maternal egg transfer. Bars represent mean 638 concentrations of sequestered toxins ± SE. Sample sizes are identical to the ones mentioned above except of n = 10 639 for L. equestris on the seed mixture with Adonis. Different letters above bars indicate significant differences among 640 treatments (p < 0.05). 16

641 642 Figure 3. Predation of lacewing larvae (C. carnea) and great tits (P. major) on three species of milkweed bugs. 643 Milkweed bugs were either raised on sunflower seeds (controls) or on seeds of the following plant species: D. 644 purpurea (H. superbus), A. vernalis (L. equestris), or C. autumnale (S. saxatilis). Please note that in the experiment 645 with S. saxatilis and lacewing larvae, sunflower raised larvae of O. fasciatus were used as control. Left panel: we 646 assessed the proportion of milkweed bug larvae that survived the first lacewing attack. (a-c): survival of H. superbus 647 (n = 20 for both diets), L. equestris (sunflower: n = 32, A. vernalis: n = 31), and S. saxatilis (sunflower: n = 20, C. 648 autumnale: n = 20). Grey (skull) is the proportion of bugs, which were killed, white is the proportion of bugs, which 649 survived the first attack by the lacewing larva. Different letters above bars indicate significant differences among 650 treatments. Right panel: Survival rates of adult milkweed bugs across three successive encounters with juvenile P. 651 major. (d-f): survival of H. superbus, L. equestris, and S. saxatilis. Only the data from bugs attacked by birds in 652 respective trials are included. Bars represent mean survival rates ± SE. 17

653 654 Figure 4. Resistance of P. apterus, O. fasciatus, and S. saxatilis to injected toxins. Adult hemipteran specimens were 655 injected with either PBS (control), the cardenolide ouabain (5 mg/ml, i.e. 5 µg/individual), or the alkaloid colchicine 656 (10 mg/ml, i.e. 10 µg/individual). On the left, we mapped resistance phenotypes on a scheme representing the 657 phylogenetic relationships of the species involved. Bar charts on the right show the percentage of individuals that 658 showed no signs of intoxication at the next day after injecting toxins (hatched). Insect icons are intended to visualize 659 either a toxic (grey bug with skull) or no toxic effect (colored bug). Numbers in stacked bars represent the actual 660 number of affected or unaffected individuals. Note that we also tested S. pandurus, a congener of S. saxatilis, and 661 found it not to possess resistance to colchicine (see Supplemental Figure 10). 662

663

664

665

666

667

668 Discussion

669 It is widely accepted that coevolution between insects and plants occurs in a multitrophic context. 670 Nevertheless, our understanding of the evolutionary drivers is still limited especially with regard 671 to the underlying mechanisms. Coevolutionary theory posits that occupation of novel dietary 672 niches depends on insect resistance to host plant toxins and that resistance traits may interfere 673 with dietary breadth. Recently, we have shown that in addition, interactions with higher trophic 674 levels (predators and parasitoids) are likely to select for specific resistance traits in insects22. 675 However, the interplay of specific adaptations and acquisition of plant toxins for defense as 676 drivers of host shifts5 and specialization has never been addressed. Here, we tested if 677 sequestration of plant toxins and preadaptation in milkweed bugs mediated specific associations 678 with particular host plants. Detailed analyses of the selective forces directing the evolution of 679 insect-plant interactions are mandatory to unravel the function of ecosystems not only for basic 680 research but also for conservation efforts and the advancement of coevolutionary theory. 681 In accordance with their global association with plants in the Apocynaceae, milkweed 682 bugs (Lygaeinae) are preadapted to sequester cardiac glycosides. We identified three milkweed 683 bug species, L. equestris, H. superbus, and S. pandurus, which independently colonized plants 684 from four botanical families (Asparagaceae, Brassicaceae, Plantaginaceae, and Ranunculaceae) 685 that produce cardiac glycosides convergently, and found that all three species sequester these 686 toxins from their evolutionarily novel hosts. A fourth species, S. saxatilis, has furthermore 687 evolved the ability to sequester alkaloids from C. autumnale (Colchicaceae), likely using some of 688 the same mechanisms for uptake, storage and release as are used for cardiac glycoside 689 sequestration. 690 Across the three systems tested (H. superbus on D. purpurea, L. equestris on A. vernalis, 691 and S. saxatilis on C. autumnale) we did not find improved growth when toxic seeds of the 692 respective novel hosts were included into seed mixtures. On pure diets of toxic A. vernalis and C. 693 autumnale seeds, growth was reduced substantially, indicating that seeds from these plant species 694 alone are not a suitable diet. Only H. superbus grew equally well on toxic D. purpurea seeds as 695 on all other diets, suggesting a higher degree of dietary specialization for this species. Our results 696 demonstrate that seeds from toxic host plants are not required for successful development, which 697 is supported by the fact that species such as H. superbus and L. equestris are easy to maintain in 698 the laboratory exclusively on sunflower seeds over many generations without an apparent effect 699 on fitness. While toxic host plants may also provide important nutritional resources temporarily, 700 it seems unlikely that specific associations with these plants were evolutionarily driven by the 701 benefit of occupying novel dietary niches. This is in line with several milkweed bug species 702 being generalist seed predators that can feed on a tremendous variety of host plant species. 703 Throughout its distributional range, L. equestris is closely associated with Vincetoxicum 704 hirundinaria, an Apocynaceae that is lacking cardiac glycosides but provides L. equestris with an 705 unknown defense29. Nonetheless, L. equestris reached smaller body size when raised on V. 706 hirundinaria compared to being raised on sunflower seeds29 and other fitness related traits such 707 as fertility, mortality, and developmental time were not different across the two diets56. In nature, 708 both L. equestris and S. saxatilis in fact use a great diversity of host plants. For L. equestris, 60 709 plant species from roughly 20 botanical families45 and for S. saxatilis more than 40 species from 710 over 15 families have been recorded (see Supplemental Table 3). Consequently, both species can 711 be considered generalists from a dietary perspective.

712 We have shown earlier that cardiac glycoside resistant Na+/K+-ATPases and sequestration 713 of cardiac glycosides apparently are synapomorphic traits of the Lygaeinae40. In addition, 714 milkweed bugs concentrate cardiac glycosides far above haemolymph levels in specialized 715 storage compartments33,34 from where they are released upon predator attack. Remarkably, we 716 found an identical suite of adaptations to colchicum alkaloids in S. saxatilis with colchicine and 717 related alkaloids being highly enriched in the defensive secretion compared to the haemolymph. 718 To tolerate sequestration of these compounds, S. saxatilis evolved a novel resistance trait against 719 colchicine that is not present in its congener S. pandurus. 720 Specialization of S. saxatilis to C. autumnale (and maybe other Colchicum species) is 721 evidenced by our screening of museum specimens. The presence of colchicum alkaloids in 30 722 randomly selected specimens from 11 countries in Europe and North Africa clearly shows that 723 each individual accessed Colchicum during its lifetime, while the lack of cardenolides suggests 724 that S. saxatilis completely shifted from the use of cardenolides to the novel defense. Oviposition 725 into Colchicum seedpods and allocation of high amounts of alkaloids into the eggs finally 726 supports a close association of S. saxatilis with C. autumnale. Remarkably, S. saxatilis still 727 maintains resistance to cardiac glycosides and accumulated resistance traits against different 728 classes of plant toxins over evolutionary time, even though target site insensitivity of Na+/K+- 729 ATPase was suggested to incur a physiological cost in O. fasciatus42. 730 The results of our predation assays revealed that feeding on either cardiac glycoside- or 731 colchicum alkaloid-containing seeds at least partially protects milkweed bugs against lacewing 732 larvae and passerine birds. We found higher survival after lacewing attacks for H. superbus 733 raised on D. purpurea seeds, and for S. saxatilis larvae that derived colchicum alkaloids from 734 their eggs. In contrast, L. equestris raised on A. vernalis seeds and S. pandurus raised on U. 735 maritima seeds were not protected against lacewings, suggesting that predator-prey interactions 736 are likely affected by the source and quality (rather than quantity) of sequestered plant toxins, the 737 sequestering insect species, or a combination of both. Nevertheless, together with previous 738 studies32,57 our results suggest that sequestration of plant toxins mediates effective defense of 739 milkweed bugs against arthropod predators. 740 In experiments with avian predators, sequestration of host plant chemicals decreased bird 741 attack rates and increased prey survival compared to sunflower-raised bugs. This effect was 742 present in all three milkweed-bug species, colchicine being equally effective as cardenolides. 743 Aversiveness of sequestered chemicals was also evidenced by discomfort-indicating behavior 744 following contact with bugs from toxic host plants. Higher effectiveness of sequestered than 745 autogenous chemicals against avian predators has also been found in other studied systems, e.g. 746 leaf beetles58 and lanternflies59. Nevertheless, our results show that the autogenous scent-gland 747 secretion alone still increases bug survival compared to undefended prey. Our findings indicate 748 that besides being highly toxic, cardenolides60 and colchicine may protect prey due to an aversive 749 taste. Consequently, the bugs derived from toxic host plants frequently survived bird attacks and 750 were almost never eaten, while sunflower-raised bugs were frequently killed, and at least partly 751 consumed. 752 The increased effectiveness of sequestration over autogenous secretion indicates that 753 milkweed bugs represent an instance of automimicry61, i.e. interspecific variation in antipredatory 754 defence when less defended or undefended individuals gain protection by resembling their better 755 defended conspecifics. Decreasing attack rates and increasing bug survival across trials suggest 756 that birds combined decisions based on visual cues with taste-sampling62,63, which allows to 757 discriminate between defended individuals and automimics64 and taste-reject only the prey that 20

758 actually contains toxins65. The reason why milkweed bugs maintain variation in host plant 759 utilization and chemical defense could be a trade-off between development and defense66 760 mediated by toxic host plants representing a suboptimal food source. Nevertheless, different 761 defense chemicals could still be equally effective against some predators29. Unpredictability of 762 defense is also considered aversive by itself67, and can increase effectiveness of avoidance 763 learning68. A broad spectrum of host plants in many milkweed-bug species suggests that in this 764 taxonomic group automimicry is evolutionary stable69,70. Intraspecific variation in antipredatory 765 defence of milkweed bugs may also affect their mimetic relationships with similarly colored prey 766 species. 767 Sequestration of cardiac glycosides from A. vernalis, D. purpurea, E. crepidifolium, and 768 U. maritima was most likely facilitated by preadaptation, yet sequestration of colchicum 769 alkaloids is an entirely novel ability. However, while colchicum alkaloids drastically differ from 770 cardiac glycosides in their mode of action or target site, the two types of compounds nonetheless 771 share some similarities: both comprise small, chemically stable, and mostly lipophilic molecules 772 that do not require enzymatic activation to become highly toxic. Therefore, even though S. 773 saxatilis likely lacked preadaptations for tolerance of the novel toxin, its aposematism and 774 sequestration machinery for uptake, specialized storage and release of toxins33 may well have 775 facilitated the evolution of colchicum alkaloid sequestration. Our findings therefore demonstrate 776 that host shifts of specialized insects can be mediated by preadaptation to specific toxins and 777 convergent evolution of plant toxins in unrelated plant taxa. At the same time, we propose that 778 suites of traits involved in sequestration of one type of chemical may similarly represent 779 preadaptations facilitating shifts to entirely novel classes of chemically unrelated compounds, 780 particularly if favored by large benefits (i.e. sequestration of highly toxic colchicine). 781 In conclusion, specialization in milkweed bugs is not an evolutionary dead end4, and 782 evolutionary plasticity is maintained by different mechanisms including preadaptation with 783 regard to different traits of the same syndrome. Our findings demonstrate that it is insufficient to 784 classify the degree of specialization in insects solely based on their trophic interactions. Species 785 that classify as dietary generalists may still specialize to host plants serving as a source of 786 sequestered toxins. Interactions driven by the third trophic level (predators and parasitoids) can 787 therefore direct specialization of bitrophic interactions between herbivores and their hostplants.

788 Acknowledgments

789 We thank Michael Falkenberg for drawing our attention to Spilostethus saxatilis emerging from 790 seedpods of Colchicum autumnale, Susanne Dobler for providing a lab strain of Oncopeltus 791 fasciatus, Andreas Berger who observed H. superbus feeding on E. crepidifolium and shared the 792 location with us, and Luis Vivas for supporting our search of S. pandurus feeding on U. 793 maritima. The Museum für Naturkunde Berlin, the Senckenberg Deutsches Entomologisches 794 Institut Müncheberg, and the Staatliches Museum für Naturkunde Karlsruhe, Germany, provided 795 dry specimens of S. saxatilis for chemical extraction. We furthermore thank all the collectors of 796 specimens and Hermann Falkenhahn for identifying host plants of S. saxatilis. We thank Anurag 797 Agrawal for comments on our manuscript. Moreover, we thank the Junta Andalucia, Spain, the 798 Landesamt für Umwelt Brandenburg, the Regierungspräsidium Karlsruhe, and the Struktur- und 799 Genehmigungsdirektion Nord, Koblenz, Germany for issuing collecting permits for bugs and 800 plant material. This work was supported by DFG grant PE 2059/3-1 to GP and the LOEWE

801 program of the State of Hesse to AV and GP via funding the LOEWE Center for Insect 802 Biotechnology & Bioresources and a Czech Science Foundation Grant 19-09323S to AE. 803

804 References

805 806 1. Ehrlich, P. R. & Raven, P. H. Butterflies and plants: a study in coevolution. Evolution (N. 807 Y). 18, 586–608 (1964).

808 2. Berenbaum, M. R. Chemical mediation of coevolution: phylogenetic evidence for 809 Apiaceae and associates. Ann. Missouri Bot. Gard. 88, 45–59 (2001).

810 3. Becerra, J. X. Insects on plants: macroevolutionary chemical trends in host use. Science 811 (80-. ). 276, 253–256 (1997).

812 4. Termonia, A., Hsiao, T. H., Pasteels, J. M. & Milinkovitch, M. C. Feeding specialization 813 and host-derived chemical defense in Chrysomeline leaf beetles did not lead to an 814 evolutionary dead end. Proc. Natl. Acad. Sci. 98, 3909–3914 (2001).

815 5. Murphy, S. M. & Feeny, P. Chemical facilitation of a naturally occurring host shift by 816 Papilio machaon butterflies (Papilionidae). Ecol. Monogr. 76, 399–414 (2006).

817 6. Dethier, V. G. Evolution of feeding preferences in phytophagous insects. Evolution (N. Y). 818 8, 33–54 (1954).

819 7. Scriber, J. M. & Feeney, P. Growth of herbivorous caterpillars in relation to feeding 820 specialization and to the growth form of their food plants. Ecology 60, 829–850 (1979).

821 8. Montandon, R., Stipanovic, R. D., Williams, H. J., Sterling, W. L. & Vinson, S. B. 822 Nutritional indices and excretion of gossypol by Alabama argillacea (Hübner) and 823 Heliothis virescens (F.) (Lepidoptera: Noctuidae) fed glanded and glandless cotyledonary 824 cotton leaves. J. Econ. Entomol. 80, 32–36 (1987).

825 9. Singer, M. S. Specialization, Speciation, and Radiation. (University of California Press, 826 2008).

827 10. Murphy, S. M. Enemy-free space maintains swallowtail butterfly host shift. Proc. Natl. 828 Acad. Sci. 101, 18048–18052 (2004).

829 11. Denno, R. F., Larsson, S. & Olmstead, K. L. Role of enemy-free space and plant quality in 830 host-plant selection by willow beetles. Ecology 71, 124–137 (1990).

831 12. Opitz, S. E. W. & Müller, C. Plant chemistry and insect sequestration. Chemoecology 19, 832 117–154 (2009).

833 13. Petschenka, G. & Agrawal, A. A. How herbivores coopt plant defenses: natural selection, 834 specialization, and sequestration. Curr. Opin. Insect Sci. 14, 17–24 (2016).

835 14. Erb, M. & Robert, C. A. M. Sequestration of plant secondary metabolites by insect 836 herbivores: molecular mechanisms and ecological consequences. Curr. Opin. Insect Sci. 837 14, 8–11 (2016).

838 15. Termonia, A., Pasteels, J. M., Windsor, D. M. & Milinkovitch, M. C. Dual chemical 839 sequestration: a key mechanism in transitions among ecological specialization. Proc. R. 840 Soc. London. Ser. B Biol. Sci. 269, 1–6 (2002).

841 16. Kelley, S. T. & Farrell, B. D. Is specialization a dead end? The phylogeny of host use in 842 Dendroctonus bark beetles (Scolytidae). Evolution (N. Y). 52, 1731–1743 (1998).

843 17. Lampert, E. C. & Bowers, M. D. Host plant influences on iridoid glycoside sequestration 844 of generalist and specialist caterpillars. J. Chem. Ecol. 36, 1101–1104 (2010).

845 18. Dyer, L. A. Tasty generalists and nasty specialists? Antipredator mechanisms in tropical 846 lepidopteran larvae. Ecology 76, 1483–1496 (1995).

847 19. Züst, T. & Agrawal, A. A. Population growth and sequestration of plant toxins along a 848 gradient of specialization in four aphid species on the common milkweed Asclepias 849 syriaca. Funct. Ecol. 547–556 (2015) doi:10.1111/1365-2435.12523.

850 20. Engler-Chaouat, H. S. & Gilbert, L. E. De novo synthesis vs. sequestration: negatively 851 correlated metabolic traits and the evolution of host plant specialization in cyanogenic 852 butterflies. J. Chem. Ecol. 33, 25–42 (2007).

853 21. Brower, L. P. Bird predation and foodplant specificity in closely related procryptic insects. 854 Am. Nat. 92, 183–187 (1958).

855 22. Petschenka, G. & Agrawal, A. A. Milkweed butterfly resistance to plant toxins is linked to 856 sequestration, not coping with a toxic diet. Proc. R. Soc. B Biol. Sci. 282, 20151865 857 (2015).

858 23. Huber, M. et al. A latex metabolite benefits plant fitness under root herbivore attack. PLoS 859 Biol 14, e1002332 (2016).

860 24. Scudder, G. G. E. & Duffey, S. S. Cardiac glycosides in the Lygaeinae (Hemiptera: 861 Lygaeidae). Can. J. Zool. 50, 35–42 (1972).

862 25. Slater, J. A. & Sperry, B. The biology and distribution of the South African Lygaeinae, 863 with descriptions of new species (Hemiptera: Lygaeidae). Ann. Transvaal Museum 28, 864 117–201 (1973).

865 26. Péricart, J. Hémiptères Lygaeidae Euro-Méditerranéens. Volume 1. Faune de France 84A. 866 (Fédération francaise des Sociétés des Sciences Naturelles, 1998).

867 27. Slater, A. A taxonomic revision of the Lygaeinae of Australia (Heteroptera: Lygaeidae). 868 Univ. Kansas Sci. Bull. 52, 301–481 (1985).

869 28. Schuh, R. T. & Slater, J. A. True Bugs of the World (Hemiptera:Heteroptera):

870 classification and natural history. (Cornell University Press, 1995).

871 29. Tullberg, B. S., Gamberale-Stille, G. & Solbreck, C. Effects of food plant and group size 872 on predator defence: differences between two co‐occurring aposematic Lygaeinae bugs. 873 Ecol. Entomol. 25, 220–225 (2000).

874 30. Evans, D. L., Castoriades, N. & Badruddine, H. Cardenolides in the defense of Caenocoris 875 nerii (Hemiptera). Oikos 46, 325–329 (1986).

876 31. McLain, D. K. & Shure, D. J. Host plant toxins and unpalatability of Neacoryphus bicrucis 877 (Hemiptera: Lygaeidae). Ecol. Entomol. 10, 291–298 (1985).

878 32. Berenbaum, M. R. & Miliczky, E. Mantids and Milkweed Bugs: efficacy of aposematic 879 coloration against invertebrate predators. Am. Midl. Nat. 111, 64–68 (1984).

880 33. Bramer, C., Friedrich, F. & Dobler, S. Defence by plant toxins in milkweed bugs 881 (Heteroptera: Lygaeinae) through the evolution of a sophisticated storage compartment. 882 Syst. Entomol. 42, 15–30 (2017).

883 34. Scudder, G. G. E., Moore, L. V & Isman, M. B. Sequestration of cardenolides in 884 Oncopeltus fasciatus: morphological and physiological adaptations. J. Chem. Ecol. 12, 885 1171–1187 (1986).

886 35. Duffey, S. S., Blum, M. S., Isman, M. B. & Scudder, G. G. E. Cardiac glycosides: a 887 physical system for their sequestration by the milkweed bug. J. Insect Physiol. 24, 639– 888 643 (1978).

889 36. Agrawal, A. A., Petschenka, G., Bingham, R. A., Weber, M. G. & Rasmann, S. Toxic 890 cardenolides: chemical ecology and coevolution of specialized plant-herbivore 891 interactions. New Phytol. 194, 28–45 (2012).

892 37. Dobler, S., Petschenka, G., Wagschal, V. & Flacht, L. Convergent adaptive evolution – 893 how insects master the challenge of cardiac glycoside-containing host plants. Entomol. 894 Exp. Appl. 157, 30–39 (2015).

895 38. Dobler, S., Dalla, S., Wagschal, V. & Agrawal, A. A. Community-wide convergent 896 evolution in insect adaptation to toxic cardenolides by substitutions in the Na,K-ATPase. 897 Proc. Natl. Acad. Sci. 109, 13040–13045 (2012).

898 39. Zhen, Y., Aardema, M. L., Medina, E. M., Schumer, M. & Andolfatto, P. Parallel 899 molecular evolution in an herbivore community. Science (80-. ). 337, 1634–1637 (2012).

900 40. Bramer, C., Dobler, S., Deckert, J., Stemmer, M. & Petschenka, G. Na+/K+-ATPase 901 resistance and cardenolide sequestration: basal adaptations to host plant toxins in the 902 milkweed bugs (Hemiptera: Lygaeidae: Lygaeinae). Proc. R. Soc. London B Biol. Sci. 282, 903 20142346 (2015).

904 41. Moore, L. V & Scudder, G. G. E. Ouabain-resistant Na,K-ATPases and cardenolide 905 tolerance in the large milkweed bug, Oncopeltus fasciatus. J. Insect Physiol. 32, 27–33 24

906 (1986).

907 42. Dalla, S. & Dobler, S. Gene duplications circumvent trade‐offs in enzyme function: insect 908 adaptation to toxic host plants. Evolution (N. Y). 70, 2767–2777 (2016).

909 43. Duffey, S. S. & Scudder, G. G. E. Cardiac glycosides in North American Asclepiadaceae, 910 a basis for unpalatability in brightly coloured Hemiptera and Coleoptera. J. Insect Physiol. 911 18, 63–78 (1972).

912 44. Von Euw, J., Reichstein, T. & Rothschild, M. Heart poisons (cardiac glycosides) in the 913 Lygaeid bugs Caenocoris nerii and Spilostethus pandurus. Insect Biochem. 1, 373–384 914 (1971).

915 45. Solbreck, C. & Kugelberg, O. Field observations on the seasonal occurrence of Lygaeus 916 equestris (L.) (Het., Lygaeidae) with special reference to food plant phenology. Insect Syst. 917 Evol. 3, 189–210 (1972).

918 46. Ollivier-Schricke, M. T. Les stratégies de reproduction de deux hétéroptères Lygaeidae. 919 Bull. Soc. Sci. Bretagne 56, 137–154 (1984).

920 47. Junior, P. & Wichtl, M. 3-epi-periplogenin: ein neues Cardenolid aus Adonis vernalis. 921 Phytochemistry 19, 2193–2197 (1980).

922 48. Vivas, L. Algunos datos sobre distribución y biología de Spilostethus furcula (Herrich- 923 Schaeffer, 1850)(Hemiptera: Heteroptera: Lygaeidae) y clave para los ligeinos ibéricos. 924 BV News, Publicaciones Cient. 1, 59–74 (2012).

925 49. Steyn, P. S. & van Heerden, F. R. Bufadienolides of plant and animal origin. Nat. Prod. 926 Rep. 15, 397–413 (1998).

927 50. Putshkov, V. Fauna Ukrainy, 21(3) Ligeidy. (Naukova Dumka, 1969).

928 51. Hotová Svádová, K., Exnerová, A., Kopečková, M. & Štys, P. How do predators learn to 929 recognize a mimetic complex: experiments with naive great tits and aposematic 930 Heteroptera. Ethology 119, 814–830 (2013).

931 52. Ravelli, R. B. G. et al. Insight into tubulin regulation from a complex with colchicine and a 932 stathmin-like domain. Nature 428, 198–202 (2004).

933 53. RCoreTeam. R: A language and environment for statistical computing. R Found. Stat. 934 Comput. Vienna, Austria (2018).

935 54. Højsgaard, S., Halekoh, U. & Yan, J. The R package geepack for generalized estimating 936 equations. J. Stat. Softw. 15, 1–11 (2006).

937 55. Vinnersten, A. & Larsson, S. Colchicine is still a chemical marker for the expanded 938 Colchicaceae. Biochem. Syst. Ecol. 38, 1193–1198 (2010).

939 56. Laukkanen, L., Kalske, A., Muola, A., Leimu, R. & Mutikainen, P. Genetic drift precluded 940 adaptation of an insect seed predator to a novel host plant in a long-term selection 25

941 experiment. PLoS One 13, e0198869 (2018).

942 57. Skow, C. D. & Jakob, E. M. Jumping spiders attend to context during learned avoidance of 943 aposematic prey. Behav. Ecol. 17, 34–40 (2005).

944 58. Zvereva, E. L. et al. Defence strategies of Chrysomela lapponica (Coleoptera: 945 Chrysomelidae) larvae: relative efficacy of secreted and stored defences against insect and 946 avian predators. Biol. J. Linn. Soc. 124, 533–546 (2018).

947 59. Song, S., Kim, S., Kwon, S. W., Lee, S.-I. & Jablonski, P. G. Defense sequestration 948 associated with narrowing of diet and ontogenetic change to aposematic colours in the 949 spotted lanternfly. Sci. Rep. 8, 16831 (2018).

950 60. Brower, L. P. & Fink, L. S. A natural toxic defense system: cardenolides in butterflies 951 versus birds. Ann. N. Y. Acad. Sci. 443, 171–188 (1985).

952 61. Ruxton, G. D., Sherratt, T. N. & Speed, M. P. Avoiding attack: the evolutionary ecology of 953 crypsis, mimicry and aposematism. Oxford Univ. Press (2004).

954 62. Skelhorn, J. & Rowe, C. Avian predators taste–reject aposematic prey on the basis of their 955 chemical defence. Biol. Lett. 2, 348–350 (2006).

956 63. Sherratt, T. N. The optimal sampling strategy for unfamiliar prey. Evolution (N. Y). 65, 957 2014–2025 (2011).

958 64. Guilford, T. ‘Go-slow’ signalling and the problem of automimicry. J. Theor. Biol. 170, 959 311–316 (1994).

960 65. Gamberale-Stille, G. & Guilford, T. Automimicry destabilizes aposematism: predator 961 sample-and-reject behaviour may provide a solution. Proc. R. Soc. London. Ser. B Biol. 962 Sci. 271, 2621–2625 (2004).

963 66. Ruxton, G. D. & Speed, M. P. How can automimicry persist when predators can 964 preferentially consume undefended mimics? Proc. R. Soc. B Biol. Sci. 273, 373–378 965 (2006).

966 67. Barnett, C. A., Bateson, M. & Rowe, C. Better the devil you know: avian predators find 967 variation in prey toxicity aversive. Biol. Lett. 10, 20140533 (2014).

968 68. Skelhorn, J. & Rowe, C. Tasting the difference: do multiple defence chemicals interact in 969 Müllerian mimicry? Proc. R. Soc. B Biol. Sci. 272, 339–345 (2005).

970 69. Speed, M. P., Ruxton, G. D. & Broom, M. Automimicry and the evolution of discrete prey 971 defences. Biol. J. Linn. Soc. 87, 393–402 (2006).

972 70. Svennungsen, T. O. & Holen, Ø. H. The evolutionary stability of automimicry. Proc. R. 973 Soc. B Biol. Sci. 274, 2055–2063 (2007).

974 71. Paine, C. E. T. et al. How to fit nonlinear plant growth models and calculate growth rates: 975 an update for ecologists. Methods Ecol. Evol. 3, 245–256 (2012). 26

976 72. Züst, T., Rasmann, S. & Agrawal, A. A. Growth–defense tradeoffs for two major anti- 977 herbivore traits of the common milkweed Asclepias syriaca. Oikos 124, 1404–1415 978 (2015).

979 73. Miles, P. W. Retention of food residues in the midgut by nymphs of the milkweed bug, 980 Oncopeltus fasciatus Dall. Nature 182, 959 (1958).

981 74. Moore, L. V & Scudder, G. G. E. Selective sequestration of milkweed (Asclepias sp.) 982 cardenolides in Oncopeltus fasciatus (Dallas) (Hemiptera: Lygaeidae). J. Chem. Ecol. 11, 983 667–687 (1985).

984 75. Luckner, M. & Wichtl, M. Digitalis. Wissenschaftliche Verlagsgesellschaft mbH Stuttgart 985 2000. (2000).

986 76. Makarevich, I. F., Klimenko, O. I. & Kolesnikov, D. G. Cardiac glycosides of Erysimum 987 crepidifolium. Chem. Nat. Compd. 10, 619–622 (1974).

988 77. Kopp, B. et al. Bufadienolides from Urginea maritima from Egypt. Phytochemistry 42, 989 513–522 (1996).

990 78. Bates, D., Mächler, M., Bolker, B. & Walker, S. Fitting linear mixed-effects models using 991 lme4. J. Stat. Softw. 67, 1–48 (2015).

992 79. Banar, P. Utilisation of toxic plants as host plants in model species of true bugs 993 (Heteroptera). (Charles University, Prague, 2003).

994 995

996

997

998

999

1000

1001

1002

1003

1004

1005

1006 Supplementary Materials

1007 Supplementary Methods

1008 Modelling of milkweed bug growth in seed mixture experiments 1009 In addition to comparing final body mass of milkweed bug larvae after three weeks of feeding, 1010 we also modelled larval growth as a continuous process using the four sequential body mass 1011 measurements recorded during the experiment (initial mass, mass at weeks 1, 2, and 3). Log- 1012 transformed body masses were modelled using an asymptotic regression model71,72, implemented 1013 as the SSasymp function in the nlme package for the statistical software R. Body mass was thus 1014 modelled as a function of the initial mass, a rate of increase, and an asymptotic mass at the end of 1015 the experiment. Effects of seed mixture were included for the rate of increase and asymptote 1016 parameters, while individual bugs (i.e. body mass means per petri dish) were treated as random 1017 effects to account for repeated measures. To compare growth between seed mixture treatments, 1018 the absolute growth rate of bugs (AGR, mass gain per day) was calculated for the final day of the 1019 experiment72.

1020 Maintenance of milkweed bugs on sunflower seeds to purge toxins from guts 1021 After the mixed seed feeding experiments, bugs were transferred to fresh sunflower seeds for 14 1022 days to clean guts from residual dietary toxins that would bias the estimate of compounds 1023 sequestered into the body tissues. Since several true bug species, including Oncopeltus fasciatus 1024 are known to have a discontinuous digestive tract until the adult stage73, we cannot rule out that 1025 larvae had toxins remaining in their guts at the time of analysis (see Supplemental Table 2 for the 1026 number of days individual specimens had for purging during the adult stage). As an estimate, we 1027 quantified colchicum alkaloids in the gut of last instar larvae of S. saxatilis (extracted in 2 x 1 ml 1028 methanol, resuspended in 100 µl methanol and analyzed as described above) that were raised on 1029 C. autumnale seeds from the second larval instar and found them to contain only 1.44 µg 1030 alkaloids per gut (n = 4, SE = 0.76). Similarly, other authors found that the amount of 1031 cardenolides in filled guts of adult O. fasciatus raised on milkweed seeds was below a detection 1032 limit of 10 µg74. Consequently, it seems unlikely that substantial amounts of non-sequestered 1033 toxins remaining in the larval gut would have biased our results.

1034 Comparison of colchicum alkaloids in haemolymph and defensive secretion of S. saxatilis 1035 To compare concentrations of sequestered colchicum alkaloids between haemolymph and 1036 defensive secretion (i.e. clear droplets released at the integument upon attack), we used adults of 1037 S. saxatilis collected in the field (Berghausen, August 2016). Before collecting samples, we 1038 maintained bugs on pure Colchicum seeds (supplied with water) for 13 days under ambient

1039 conditions. To obtain defensive secretion, we anaesthetized bugs with CO2 and squeezed them 1040 between the blades of broad tweezers. We collected emerging droplets of clear defensive fluid 1041 using 0.5 µl glass capillaries. Next, we cut off a hind-leg at the femur to collect haemolymph. We 1042 determined the collected volume based on the filling level of the capillaries. In total, we obtained 1043 13 samples of haemolymph and seven samples of defensive secretion (including six paired 1044 samples, i.e. from the same individual). One haemolymph sample was excluded since the 1045 colchicum alkaloid peaks were too small for automatic integration. Before HPLC-analysis, filled 1046 capillaries were stored at -80°C until extraction. To analyze alkaloids, whole capillaries including 1047 liquid samples were homogenized and extracted with methanol as described in the HPLC- 28

1048 methods section of the main manuscript. Statistical comparison of colchicum alkaloid 1049 concentrations between S. saxatilis defensive secretion and haemolymph was carried out using 1050 matched pair analysis based on six paired samples (i.e., six samples of haemolymph and six 1051 samples of secretions from the same individuals).

1052 HPLC analysis of museum specimens 1053 To test for the occurrence of colchicum alkaloids and cardiac glycosides in museum specimens of 1054 S. saxatilis, we incubated dry insects in 1 ml of methanol for at least one day and collected 1055 supernatants in fresh vials. After repeating this procedure twice (i.e., 3 ml methanol in total), 1056 pooled supernatants were evaporated under N2 and dissolved in 100 µl of methanol using a 1057 FastPrep homogenizer (6.5 m/s, 45 sec.). Before analyzing samples for colchicum alkaloids or 1058 cardenolides via HPLC using the respective methods (see above), samples were centrifuged 1059 (16,100 x g, 3 min) and filtered as described above.

1060 Preparing seeds for HPLC 1061 To compare toxins from host plant seeds to toxins sequestered by the bugs, seeds of A. vernalis 1062 (10.51 – 14.47 mg, commercial source), C. autumnale (5.91 – 10.03 mg, Berghausen, Germany, 1063 2016), D. purpurea (59.44 – 63.72 mg, Eberbach, Germany, 2017), E. crepidifolium (10.19 – 1064 10.65 mg, Schloßböckelheim, Germany, 2018), and U. maritima (2.89 – 7.66 mg, Aracena, 1065 Spain, 2016) were weighed and extracted with 0.5 ml (A, vernalis, C. autumnale, U. maritima) or 1066 1 ml (D. purpurea, E. crepidifolium) methanol containing 0.01 mg/ml digitoxin (A. vernalis) or 1067 oleandrin (D. purpurea) as an internal standard (no internal standard for C. autumnale, E. 1068 crepidifolium, and U. maritima). For homogenization in a FastPrep homogenizer (2 x 6.5 m/s, 45 1069 s), we used either zirconia beads for D. purpurea and E. crepidifolium or lysing matrix A (MP 1070 Biomedicals) for A. vernalis, C. autumnale and U. maritima. After grinding, samples were 1071 centrifuged (16,100 x g, 3 min) and supernatants were transferred to fresh vials. This procedure 1072 was repeated once for D. purpurea and E. crepidifolium seeds and twice for A. vernalis, C. 1073 autumnale, and U. maritima seeds. Supernatants of individual samples were pooled and 1074 evaporated under N2. Before HPLC-analyses, dried residues were dissolved in 500 (C. 1075 autumnale, U. maritima), 200 (D. purpurea, E. crepidifolium) or 100 µl methanol (A. vernalis) 1076 and analyzed as described above.

1077 Evaluation of chromatograms 1078 For chromatograms obtained from extracts of field-collected H. superbus from D. purpurea, L. 1079 equestris, S. pandurus, and S. saxatilis we considered all peaks of a sample showing compound 1080 specific absorption spectra (i.e. cardenolides, bufadienolides, and colchicum alkaloids) unless 1081 peaks were not automatically recognized by the software or had a signal to noise ratio < 2:1. For 1082 eggs of L. equestris and S. saxatilis as well as for the comparison between S. saxatilis secretion 1083 and haemolymph, the same approach was used. For samples obtained during seed mixture assays, 1084 we followed the procedure described above for S. saxatilis. Peaks in chromatograms from 1085 extracts of H. superbus and L. equestris were only taken into account if they were present in at 1086 least 70% or 60% of samples, respectively. The same approach was used for H. superbus, L. 1087 equestris larvae obtained from the feeding experiments with lacewings. For larvae of S. saxatilis 1088 and S. pandurus peaks were included that were present in at least 80% or 70% of samples, 1089 respectively. To evaluate field-collected H. superbus from E. crepidifolium we included all peaks

1090 that occurred in at least in 65% of samples. All datasets that were compared statistically were 1091 evaluated based on identical criteria.

1092 Structural identification via reference compounds and liquid chromatography – mass 1093 spectrometry 1094 To verify structural identity of selected cardenolides and colchicum alkaloids we compared 1095 HPLC retention times of chromatographic peaks obtained from field-collected L. equestris 1096 samples with authentic standards of k-strophanthoside (Roth, Germany), strophanthidin 1097 (PhytoLab, Germany) and cymarin (PhytoLab, Germany), cardenolides which are known to 1098 occur in A. vernalis47. For the same purpose, extracts from H. superbus were compared to 1099 authentic digitoxigenin (Sigma-Aldrich, Germany), digitoxin (Sigma-Aldrich, Germany), 1100 digoxigenin (Sigma-Aldrich, Germany), digoxin (Sigma-Aldrich, Germany), gitoxigenin (Santa 1101 Cruz Biotechnology, USA), gitoxin (EDQM, France), lanatoside C (Sigma-Aldrich, Germany), 1102 purpurea glycoside A (EDQM, France), and purpurea glycoside B (EDQM, France) which are 1103 known to occur in D. purpurea75, and erysimoside (Latoxan, France), known to occur in E. 1104 crepidifolium76. Similarly, we screened extracts of S. pandurus collected from U. maritima for 1105 the Urginea bufadienolides proscillaridin A (PhytoLab, Germany) and scillaren A (Sigma- 1106 Aldrich, Germany)77(Supplemental Figure 2e). Last, we compared chromatograms from field- 1107 collected S. saxatilis to the authentic colchicum alkaloids colchicoside, 2-demethylcolchicine, 3- 1108 demethylcolchicine (Toronto Research Chemicals, Canada), and colchicine (Roth, Germany) 1109 (Supplemental Figure 2e). In addition to the comparison of HPLC retention times, we compared 1110 mass spectra of the colchicum alkaloid compounds in the bug extracts to spectra of authentic 1111 alkaloid standards. Mass spectral analyses were performed on a Bruker micrOTOF-Q II mass 1112 spectrometer equipped with a Dionex Ultimate 3000 UHPLC and a C18 HPLC column (Kinetex 1113 C18, 2.6µ, 100A, 150 x 2.1 mm; Phenomenex). Compounds were separated by gradient elution 1114 with a flow rate of 150 mL min-1 and the following gradient of solvent A (0.1% formic acid in 1115 water) and solvent B (0.1% formic acid in acetonitrile): 0 to 2 min 16% B, 25 min 70% B, 30 min 1116 95%. The column was eluted with 95% B solvent for an additional 10 min and re-equilibrated at 1117 the starting condition for 5 min. Sodium formate infusions at the beginning of each sample run 1118 were used for the mass calibration.

1119 Effect of sequestered toxins on consumption of milkweed bug larvae by lacewing larvae 1120 Milkweed bug and lacewing larvae were weighed before the trial and after exposition of bugs to 1121 lacewing larvae (after 12 hours for H. superbus and 8 hours for the other species) to assess 1122 consumption of milkweed bug larvae by lacewings. For statistical analysis, only dead bugs and 1123 corresponding lacewing larvae were included to ensure that milkweed bug larvae had been 1124 actually attacked by the lacewings. The aim of this experiment was to evaluate potential 1125 deterrence of sequestered plant compounds on lacewing feeding. 1126 To assess the effect of sequestered toxins on remaining body mass of milkweed bug 1127 larvae after partial consumption by lacewings, remaining body mass was log10-transformed. We 1128 tested potential differences between dietary treatments using ANCOVA in JMP and included the 1129 initial mass of the intact milkweed bug larvae (i.e. before the experiment) as well as the initial 1130 mass of the lacewing larvae (as an estimate for body size) in our model. Furthermore, we tested 1131 for an interaction between the dietary treatment and the initial mass of the bugs. Since we carried 1132 out two experimental rounds for L. equestris, we included ‘experiment’ as a blocking term. For 1133 the remains of one out of 52 L. equestris individuals and four out of 40 S. saxatilis individuals, 30

1134 body mass was zero (i.e. the mass was below the detection limit of the balance used), thus these 1135 data were removed before log10-transformation. 1136 For the comparison of lacewing weights after feeding on milkweed bug larvae, we used 1137 the same model as for milkweed bugs using untransformed data with final mass of the lacewings 1138 as the main effect. We included the initial weight of bugs and lacewings as model effects. Due to 1139 the two experimental rounds for L. equestris, we again included ‘experiment’ as a blocking term. 1140 Initial masses of milkweed bugs and lacewings across treatments (i.e. bugs raised on sunflower 1141 vs. toxic seeds) were compared by two-tailed Student’s t-tests in JMP except of the experiment 1142 with L. equestris that we analyzed with an ANCOVA again using ‘experiment’ as a blocking 1143 term due to the two experimental rounds for this species.

1144 Hand rearing juvenile birds for predation assays 1145 Juvenile great tits were obtained from a population breeding in nest-boxes in mixed woods at the 1146 outskirts of Prague. The juveniles were taken from nest-boxes when 12–15 days old, and hand 1147 reared in the laboratory. This way, they were naive in regard to experience with any kind of 1148 unpalatable or warningly colored prey. The birds were kept in artificial nests until fledging and 1149 then housed in groups of three or four in indoor cages (60 x 50 x 50cm) under illumination and 1150 temperature regimes simulating natural conditions. Their diet consisted of mealworms (Tenebrio 1151 molitor larvae) and commercial mixtures for hand-rearing passerine birds (Handmix, NutriBird, 1152 Gold Patee and Uni Patee Premium, Orlux). Birds were used for predations assays when at least 1153 35-days old and fully independent. We ringed the birds individually and released them back to 1154 the locality of capture within a few days after experimentation. For experiments with great tits, 1155 we obtained permissions from the Environmental Department of Municipality of Prague (S- 1156 MHMP-83637/2014/OZP-VII-3/R-8/F), Ministry of Agriculture (13060/2014-MZE-17214), and 1157 Ministry of the Environment of the Czech Republic (42521/ENV/14-2268/630/14).

1158 Attack latencies and duration of discomfort-indicating behavior in avian predators 1159 In each trial of behavioral assays with great tits as predators, we recorded latency of the 1160 first attack and duration of discomfort-indicating behavior observed in birds (beak wiping and 1161 head shaking). Data were log10-transformed to meet the assumptions of normality and 1162 homogeneous variance and analyzed in the lme478 and geepack54 packages in R53. 1163 Attack latencies in the first trial were analyzed using a general linear model (ANOVA) 1164 with bug species and host plant toxicity entered as fixed effects. Latencies of the first attacks 1165 were also compared between bugs and control crickets by paired t-test. Changes in attack 1166 latencies over the three trials were analyzed using a generalized estimating equation model (GEE) 1167 in a subset of data including only the birds that attacked all three bugs offered. Trial number, bug 1168 species and host plant toxicity were entered as fixed effects and bird individual (id) as a random 1169 effect. 1170 Durations of discomfort-indicating behavior recorded during the first trial were analyzed 1171 using a general linear model (ANOVA) with bug species and host plant toxicity entered as fixed 1172 effects. To evaluate whether the general effect of host plant toxicity on discomfort-indicating 1173 reactions of birds also holds for each of the milkweed bug species studied, similar models were 1174 run separately for each bug species. 1175 1176

1177 Survival of milkweed bugs compared to control palatable prey 1178 We used GEE models (package geepack54 in R53) to compare survival rates of milkweed bugs 1179 raised on sunflower with survival rates of control crickets. The data included only the birds that 1180 attacked all three bugs offered and were analyzed for each milkweed bug species separately. Prey 1181 (cricket versus bug) and trial number were entered as fixed effects and bird individual (id) as a 1182 random effect.

1183 Analysis of bugs eaten by avian predators 1184 Following the observations of birds consuming the bugs in particular trials, we examined the 1185 remaining parts of the bugs using a stereomicroscope to determine what body parts the birds were 1186 able to consume. In a subset of data from first trials including only the cases when the bugs were 1187 killed, we analyzed an effect of host plant on the probability that at least part of the bug would be 1188 consumed. The data were analyzed separately for each species by generalized linear models 1189 (GLM) with binomial errors using the lme4 package78 in R53, and the hostplant was entered as a 1190 fixed effect. In addition, we compared frequency of different body parts of bugs consumed by the 1191 birds using Fisher’s exact test.

1192 Supplementary Results

1193 Developmental time on different diets 1194 H. superbus developed equally fast on all diets (F3,35 = 2.22, p = 0.103, n = 9 for the seed mixture 1195 without Digitalis and n = 10 for all other treatments) while the dietary treatment affected 1196 developmental time in L. equestris (F3,40 = 7.168, p < 0.001, n = 11 for all diets). Larvae needed 1197 longer to reach the adult stage on A. vernalis seeds compared to sunflower seeds and the seed 1198 mixture containing A. vernalis seeds (LSMeans Tukey HSD: p = 0.002; p = 0.001) but not 1199 compared to the A. vernalis-free seed mixture (LSMeans Tukey HSD: p = 0.097). Growth 1200 between the other diets was not different (LSMeans Tukey HSD: seed mixture vs. seed mixture 1201 with A. vernalis seeds, p = 0.3; seed mixture vs. sunflower, p = 0.48; sunflower vs. seed mixture 1202 with A. vernalis, p = 0.987).

1203 Sequestration on toxic seeds across different diets 1204 On seed mixtures containing low amounts of toxic seeds of either Digitalis, Adonis, or 1205 Colchicum, sequestration of plant toxins was reduced by approximately 50 % in H. superbus 1206 (Welch’s test: F1,19 = 18.863, p < 0.001) and S. saxatilis (Welch’s test: F3,20 = 83.568, p < 0.001; 1207 Games-Howell post-hoc test significant at p < 0.05 for all comparisons except of pure sunflower 1208 seeds and the seed mixture without C. autumnale seeds) while L. equestris accumulated only 1209 about 10 % of the amount of cardenolides it contained on the pure toxic diet (Welch’s test: F1,10 = 1210 66.206, p < 0.001, (Figure 2a-c). Notably, we found toxins in all (n = 32) specimens sampled 1211 from the seed mixture treatments including D. purpurea, A. vernalis, or C. autumnale seeds, 1212 showing that toxic seeds were always accessed by the bugs, even if only available in small 1213 numbers within mixtures.

1214 Natural history remarks 1215 When collecting milkweed bugs in the field we also observed feeding behavior and host plant 1216 use. In general, feeding was mainly restricted to fruits or flowers. H. superbus was only observed 1217 on D. purpurea plants and substrates such as bark and stumps but never on other plants. Early in

1218 the season, we found the bugs walking on Digitalis-leaves and stems or feeding on flowers. Later 1219 in the season, we typically found H. superbus in the opened ripe Digitalis-pods (adults and 1220 larvae). In a Digitalis-free habitat, we found H. superbus exclusively on seedpods of E. 1221 crepidifolium. In this habitat, the insects may also suck on the fleshy leaves of Sedum spec., a 1222 cushion plant that they may use as a refuge. The feeding ecology of L. equestris has been 1223 described elsewhere45. Our own observations confirm that in A. vernalis habitats, Adonis is the 1224 primary host plant early in the season. We and others recorded S. saxatilis feeding on more than 1225 40 plant species from more than 15 botanical families (Supplemental Table 3). S. saxatilis 1226 oviposits into C. autumnale seedpods and early larval stages of S. saxatilis were only observed in 1227 ripe fruits of C. autumnale (Supplemental Figure 6).

1228 Effects of sequestered Digitalis-cardenolides on consumption of H. superbus by lacewing larvae 1229 Although individuals of H. superbus raised on sunflower seeds had a greater body mass 1230 compared to individuals raised on Digitalis seeds before the experiment (t = 2.822, df = 21, p = 1231 0.01, n = 11 for sunflower-raised and n = 12 for D. purpurea-raised bugs), carcasses from 1232 sunflower-raised bugs were lighter after attack compared to carcasses from Digitalis raised bugs 1233 (F1,18 = 39.205, p < 0.001, Supplemental Figure 7a). The body mass of bugs prior to lacewing 1234 feeding (‘initial mass’) affected the remaining mass (F1,18 = 23.825, p < 0.001), while initial mass 1235 of lacewings (i.e. before feeding on a milkweed bug larva) had no effect (F1,18 = 2.636, p = 1236 0.122). There was no interaction between initial mass of bugs and treatment (F1,18 = 0.269, p = 1237 0.611) on remaining mass. In accordance with lower consumption of Digitalis-raised bugs, 1238 lacewing larvae gained more body mass when feeding on sunflower raised-bugs compared to 1239 feeding on Digitalis-raised bugs (F1,18 = 48.833, p < 0.001, Supplemental Figure 7e). The initial 1240 masses of bugs and lacewings influenced the final lacewing mass (F1,18 = 9.091, p = 0.007; F1,18 = 1241 306.9, p < 0.001). In addition, final mass of lacewings was affected by an interaction between the 1242 initial mass of bugs and their diet (F1,18 = 7.258, p = 0.015). The initial mass of lacewing larvae 1243 before the trial was not different (t = -0.059, df = 21, p = 0.953). 1244 1245 Effects of sequestered Adonis-cardenolides on consumption of L. equestris by lacewing larvae. 1246 The initial mass of bugs (analyzed by ANCOVA to account for experimental blocks, n = 21 for 1247 sunflower-raised and n = 30 for A. vernalis-raised bugs) neither differed between diets (F1,48 = 1248 1.026, p = 0.316) nor between the two experimental rounds (F1,48 = 0.046, p = 0.831). 1249 Consumption of milkweed bugs by lacewing larvae was higher on sunflower-raised bugs 1250 compared to bugs raised on Adonis (F1,45 = 9.170, p = 0.004, Supplemental Figure 7b). Again, the 1251 remaining mass of bugs was affected by their inital mass (F1,45 = 23.854, p < 0.001) and there 1252 was no interaction between initial mass and treatment (F1,45 = 0.235, p = 0.630). The mass of 1253 lacewing larvae before the trial was not different across diets and experiments (F1,49 = 0.087, p = 1254 0.77; F1,49 = 0.953, p = 0.334; n = 22 for lacewings feeding on sunflower and n = 30 for 1255 lacewings feeding on A. vernalis-raised bugs) but predicted how much body mass of the bugs 1256 was consumed (F1,49 = 6.286, p = 0.016). In accordance with the observed loss of body mass in 1257 the bugs, lacewings gained more body mass when feeding on sunflower-raised bugs compared to 1258 Adonis-raised bugs (F1,46 = 8.02, p = 0.007, Supplemental Figure 7f). The initial masses of bugs 1259 and lacewings determined the final body mass of the lacewing larvae (F1,46 = 6.286, p = 0.016; 1260 F1,46 = 98.245, p < 0.001). While the interaction between diet and initial mass of the bugs had an 1261 effect on lacewing final mass the round of experiment had no influence (F1,46 = 4.3, p = 0.044; 1262 F1,46 = 0.045, p = 0.833). 33

1263 Effects of sequestered Urginea-bufadienolides on consumption of S. pandurus by lacewing 1264 larvae. Initial mass of bugs did not differ between the two diets (sunflower vs. Urginea seeds, t = 1265 -0.173, df = 36, p = 0.864, n = 19 each). In contrast to the other species, we only found a 1266 marginal effect of the diet on remaining milkweed bug mass after the lacewing attack (F1,33 = 1267 3.79, p = 0.06, Supplemental Figure 7c). The initial mass of the bugs affected the remaining mass 1268 after the experiment (F1,33 = 21.202, p < 0.001) and there was a significant interaction between 1269 diet and initial mass affecting remaining body mass (F1,33 = 5.602, p = 0.024). The initial mass of 1270 lacewing larvae had no effect on the remaining mass of the bugs after consumption (F1,33 = 1.199, 1271 p = 0.282). Nevertheless, lacewing larvae were heavier after feeding on sunflower raised S. 1272 pandurus larvae compared to feeding on bugs raised on Urginea seeds (F1,33 = 195.969, p < 1273 0.001, Supplemental Figure 7g). Besides the diet, the initial mass of bugs and lacewings as well 1274 as the interaction between initial mass of bugs and diet affected body mass of lacewings after the 1275 trial (F1,33 = 44.982, p < 0.001; F1,33 = 174.744, p < 0.001; F1,33 = 20.639, p < 0.001). The mass of 1276 lacewing larvae was not different across treatments before the experiment (t = 1.052, df = 32, p = 1277 0.301). 1278 1279 Effects of sequestered Colchicum-alkaloids on consumption of S. saxatilis by lacewing larvae. 1280 Initial mass was higher in Oncopeltus (n = 16) than in S. saxatilis (n = 20; t = 12.533, df = 30, p < 1281 0.001, Supplemental Figure 7d) but after lacewing predation, the pattern was reversed (F1,31 = 1282 57.516, p < 0.001). The initial masses of bugs and lacewings did not affect remaining mass of 1283 bugs after attack (F1,31 = 2.91, p = 0.098 and F1,31 = 0.012, p = 0.914) and there was no 1284 interaction between the initial mass of bugs and treatment affecting remaining body mass after 1285 attack (F1,31 = 0.143, p = 0.708). Lacewing larvae assigned to O. fasciatus (n = 20) larvae were 1286 heavier compared to lacewing larvae assigned to larvae of S. saxatilis (n = 20; t = 4.324, df = 38, 1287 p < 0.001). Nevertheless, this difference was more pronounced after the trial (2-fold vs. 1.75-fold, 1288 F1,35 = 129.881, p < 0.001, Supplemental Figure 7h). The initial masses of bugs and lacewings 1289 affected the final mass of lacewings (F1,35 = 12.225, p = 0.001; F1,35 = 801.952, p < 0.001) and 1290 there was an interaction between the initial mass of bugs and treatment affecting lacewing final 1291 mass (F1,35 = 41.971, p < 0.001).

1292 Attack latency and duration of discomfort-indicating behavior in avian predators 1293 In first encounters with novel prey, birds did not hesitate longer before attacking milkweed bugs 1294 than before attacking control crickets (paired t-test, t = 0.173, df = 99, p = 0.863; Supplemental 1295 Figure 8a). Furthermore, attack latencies were influenced neither by milkweed-bug species 1296 (ANOVA, F2,94 = 1.241, p = 0.294) nor by host plant toxicity (ANOVA, F1,94 = 1.876, p = 0.117). 1297 Thus, we have not found any evidence of an innate bias against warningly coloured milkweed 1298 bugs in juvenile great tits or for before-attack defensive effects of sequestered host plant toxins. 1299 Nevertheless, birds that attacked all three bugs in a row hesitated longer before attacking the bugs 2 1300 raised on toxic host plants upon repeated encounters (GEE, trial, χ 1 = 0.931, p = 0.335; host 2 2 1301 plant, χ 1 = 27.935, p ˂ 0.001; trial: host plant interaction, χ 1 = 12.384, p ˂ 0.001; Supplemental 1302 Figure 8b). 1303 When handling the bugs, birds often responded by discomfort-indicating behavior (head 1304 shaking and beak wiping). Duration of this behavior in the first trial was similar across all bug 1305 species tested (ANOVA, F2,94 = 1.928, p = 0.151). However, the birds tested with bugs raised on 1306 toxic host plants spent more time by discomfort–indicating behavior than the birds tested with 1307 bugs from control sunflower (ANOVA, F1,94 = 37.980, p ˂ 0.001; Supplemental Figure 9). There 34

1308 was no interaction between the two factors (ANOVA, F2,94 = 0.987, p ˂ 0.001 = 0.376). When 1309 analyzed separately for each species, birds spent longer time by discomfort-indicating behavior 1310 when attacking and handling bugs raised on toxic host plants than when handling bugs from 1311 sunflower (ANOVA, S. saxatilis: F1,38 = 11.271, p = 0.002, L. equestris: F1,38 = 20.962, p ˂ 1312 0.001, H. superbus: F1,18 = 6.177, p = 0.023; Supplemental Figure 9a-c). This result indicates that 1313 chemicals sequestered from host plants cause stronger aversion in avian predators upon direct 1314 contact than the secretion of metathoracic glands alone.

1315 Survival of milkweed bugs compared to control palatable prey 1316 In all three species of milkweed bugs, the probability to survive repeated attacks by avian 2 1317 predators was higher for the sunflower-raised bugs than for control crickets (GEE, S. saxatilis: χ 1 2 2 1318 = 6.235, p = 0.012; L. equestris: χ 1 = 6.534, p = 0.011; H. superbus: χ 1 = 3.899, p = 0.048). 1319 These results indicate that the line of defence based on the secretion of metathoracic scent glands 1320 is effective by itself, even though its sole effect is considerably smaller compared to when it is 1321 combined with sequestration of host plant chemicals.

1322 Consumption of bugs by avian predators 1323 In all three milkweed bug species tested, host plant affected the probability that the birds would 2 1324 eat at least a part of the bug attacked and killed in the first trial (GLM, S. saxatilis: χ 1,24 = 2 2 1325 13.214, P ˂ 0.001; L. equestris: χ 1,27 = 5.849, P = 0.016; H. superbus: χ 1,15 = 7.945, P = 0.005). 1326 Whereas the birds frequently ate at least some parts of sunflower-raised bugs, consumption of 1327 bugs raised on toxic host plants was exceptionally rare. Out of 20 birds tested with S. saxatilis 1328 and L. equestris, only two and four birds, respectively, ate some parts of Colchicum- and Adonis- 1329 raised bugs, and in all cases it was only a small part of the abdomen (fat body). Likewise, out of 1330 10 birds tested with H. superbus, only one bird consumed a small part of the abdomen of a 1331 Digitalis-raised bug. Some of the birds tested with bugs from non-toxic hostplants consumed 1332 whole bugs or left only few fragments of cuticle, but they nevertheless consumed parts of the 1333 abdomen (usually the fat body) significantly more frequently than other parts of the bug (two 1334 tailed Fisher’s exact test, S. saxatilis: P ˂ 0.001; L. equestris: P ˂ 0.001; H. superbus: P ˂ 0.021).

1335 Structural identification of sequestered compounds and comparison to seed extracts 1336 We were not able to identify individual compounds by comparing insect extracts to nine 1337 cardenolide reference compounds known to occur in D. purpurea75 (Supplemental Figure 2a). 1338 Similarly, the cardenolide profile from D. purpurea seeds revealed only little putative overlap 1339 with cardenolides sequestered by the insects (Supplemental Figure 3a) indicating extensive 1340 metabolic transformation by the insect. In contrast, comparison of H. superbus collected from E. 1341 crepidifolium revealed clear overlap with seed extracts (Supplemental Figure 3d) and we 1342 identified one substance as erysimoside based on the retention time of a commercial reference 1343 compound (Supplemental Figure 2d). The comparison of seed extracts from A. vernalis to 1344 extracts of L. equestris revealed no clear similarity (Supplemental Figure 3b). Three cardenolides 1345 sequestered by L. equestris were putatively identified as k-strophanthosid, strophanthidin and 1346 cymarin based on authentic reference standards (Supplemental Figure 2b). The comparison of the 1347 HPLC profiles between insect extracts and Urginea seeds suggested that at least the dominant 1348 bufadienolides found in the insect are identical to compounds found in the seeds (Supplemental 1349 Figure 3c). Nevertheless, we did not detect the bufadienolides scillaren A and proscillaridin A 1350 (Supplemental Figure 2e) that are reported to occur in bulbs of U. maritima77.

1351 In S. saxatilis from Berghausen we identified 12 peaks with absorption spectra similar to 1352 colchicine. Three of the four dominant peaks present in all extracts and accounting for > 90 % of 1353 the observed alkaloids were identified as colchicine (34.6 %), 2-demethyl-colchicine (16.2 %), 1354 and 3-demethyl-colchicine (20.1 %) based on retention time and molecular mass using authentic 1355 standards (Supplemental Figure 2c). We furthermore identified the colchicine glycoside 1356 colchicoside as a minor compound. In specimens obtained from Nüstenbach, we detected only 1357 nine peaks eight of which were identical with the ones observed in S. saxatilis from Berghausen 1358 including colchicine, 2 and 3-demethyl-colchicine, and colchicoside. Seeds of C. autumnale 1359 contained colchicine and colchicoside but none of the related alkaloids found in the bugs 1360 (Supplemental Figure 3e).

1361 Injection experiments with ouabain and colchicine 1362 P. apterus showed clear signs of intoxication in a dose-dependent manner for both toxins, 1363 ouabain (Cochrane-Armitage trend test, Z = -4.006, p < 0.001) and colchicine (Cochrane- 1364 Armitage trend test, Z = -4.887, p < 0.001, Supplemental Figure 10a,d). While unaffected by 1365 ouabain (Figure 4), O. fasciatus responded to colchicine in a dose-dependent manner (Cochrane- 1366 Armitage trend test, Z = -3.453, p < 0.001, Supplemental Figure 10b). At a dose of 5 µg per 1367 animal, 100% of individuals were affected while S. saxatilis tolerated up to 30 µg which was the 1368 highest dose tested (p = 0.467 compared to specimens injected with 5 µg ouabain that were all 1369 surviving; two tailed Fisher’s exact test, Supplemental Figure 10e). S. pandurus, although being a 1370 congener of S. saxatilis, was unable to tolerate colchicine and was affected in a dose dependent 1371 fashion (Cochrane-Armitage trend test, Z = -3.933, p < 0.001, Supplemental Figure 10c).

1372

1373 Supplemental Table 1. Plant species used for seed mixture experiments. Botanical families in parentheses. For S. 1374 saxatilis and H. superbus mean weights of seeds per plant species ranged from 12 - 13 mg (with 91 % of plant 1375 species being close to 12 mg) and from 9 - 13 mg for L. equestris, with 87 % of host seed species being close to 9 mg 1376 per Petri dish. Deviations from 9 mg are due to the extensive seed sizes of certain species (T. pratensis) i.e. 1377 individual seeds exceeded 9 mg. 1378 Plant species used for S. saxatilis Plant species used for H. superbus Plant species used for L. equestris Achillea millefolium (Asteraceae) Achillea millefolium (Asteraceae) Achillea millefolium (Asteraceae) Bupleurum falcatum (Apiaceae) Bupleurum falcatum (Apiaceae) Angelica archangelica (Apiaceae) Centaurea jacea (Asteraceae) Centaurea jacea (Asteraceae) Centaurea scabiosa (Asteraceae) Cichorium intybus (Asteraceae) Cichorium intybus (Asteraceae) Cichorum intybus (Asteraceae) Daucus carota (Apiaceae) Daucus carota (Apiaceae) Cirsium arvense (Asteraceae) Hieracium pilosella (Asteraceae) Hieracium pilosella (Asteraceae) Dactylis glomerata (Poaceae) Origanum vulgare (Lamiaceae) Origanum vulgare (Lamiaceae) Daucus carota (Apiaceae) Plantago major (Plantaginaceae) Plantago major (Plantaginaceae) Hieracium pilosella (Asteraceae) Tanacetum vulgare (Asteraceae) Tanacetum vulgare (Asteraceae) Pimpinella saxifraga (Apiaceae) Taraxacum officinale (Asteraceae) Taraxacum officinale (Asteraceae) Plantago major (Plantaginaceae) Tragopogon pratensis (Asteraceae) Tragopogon pratensis (Asteraceae) Taraxacum officinale (Asteraceae) Thymus serpyllum (Lamiaceae) Tragopogon pratensis (Asteraceae) Urtica dioica (Urticaceae) Vincetoxicum hirundinaria (Apocynaceae) 1379

1380 Supplemental Table 2. Number of days bugs were spending on sunflower seeds as adults (i.e. with a continuous gut 1381 lumen) for purging digestive tracts from potential residues of dietary toxins (cardenolides or colchicum alkaloids).

S. saxatilis L. equestris H. superbus mix with Colchicum pure Colchicum mix with Adonis pure Adonis mix with Digitalis pure Digitalis (n = 11) (n = 11) (n = 10) (n = 11) (n = 11) (n = 10) 0 0 12 14 0 7 3 0 14 3 10 7 1 0 14 7 11 11 1 0 7 0 8 8 3 0 8 8 10 0 3 0 14 0 11 4 3 1 14 0 4 3 6 0 14 6 7 10 3 0 11 6 8 6 1 0 14 0 11 11 4 0 6 7 11 1382

1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406

1407 Supplemental Table 3. Host plants recorded for S. saxatilis in the field. Superscript numbers indicate sources of 1408 host plant data: 1 = own observation, 2 = Péricart (1998)26, 3 = Banar (2003)79. Asterisks indicate actual feeding 1409 observations. Plant species Location Plant family Bupleurum falcatum Germany, Nüstenbach*,1 Apiaceae Daucus carota Germany, Nüstenbach*,1 Apiaceae Heracleum Germany2 Apiaceae Asclepias syriaca ‘Europe orientale’2 Apocynaceae Vincetoxicum hirundinaria ‘Europe orientale’2, Czech Republic, South Apocynaceae Moravia*,3 Vincetoxicum stepposum ‘Europe orientale’2 Apocynaceae Achillea millefolium Czech Republic, South Moravia*,3 Asteraceae Carduus acanthoides Czech Republic, South Moravia*,3 Asteraceae Centaurea scabiosa Czech Republic, South Moravia*,3 Asteraceae Centaurea spec. Germany, Nüstenbach*,1 Asteraceae Cichorium intybus Germany, Nüstenbach(*),1, Czech Republic, South Asteraceae Moravia*,3 Cirsium pannonicum Czech Republic, South Moravia*,3 Asteraceae Cirsium spec. Germany, Nüstenbach1 Asteraceae Cirsium vulgare Czech Republic, South Moravia*,3 Asteraceae Crepis biennis Germany, Berghausen*,1 Asteraceae Hieracium Germany2 Asteraceae Inula conyzae Germany, Nüstenbach*,1 Asteraceae Inula hirta Czech Republic, South Moravia*,3 Asteraceae Leontodon hispidus Germany, Berghausen(*),1 Asteraceae Leucanthemum vulgare Czech Republic, South Moravia*,3 Asteraceae Senecio cf. vulgaris1 (stem, flower buds) Germany, Berghausen*,1 Asteraceae Senecio jacobaea Czech Republic, South Moravia*,3 Asteraceae Solidago virgaurea Germany, Nüstenbach(*),1 Asteraceae Tanacetum vulgare Czech Republic, South Moravia*,3 Asteraceae Taraxacum sect. Ruderalia Czech Republic, South Moravia*,3 Asteraceae Tragopogon Germany2 Asteraceae Colchicum autumnale Germany, Berghausen*,1, Nüstenbach*,1, Austria1,2, Colchicaceae Czech Republic, South Moravia*,3 Knautia arvensis Germany, Berghausen(*),1 Dipsacaceae Euphorbia cyparissias Czech Republic, South Moravia*,3 Euphorbiaceae Ononis spinosa Czech Republic, South Moravia*,3 Fabaceae Trifolium repens Czech Republic, South Moravia*,3 Fabaceae Hypericum perforatum Czech Republic, South Moravia*,3 Hypericaceae Mentha longifolia Italy2, Eastern Europe2 Lamiaceae Origanum vulgare Germany, Nüstenbach*,1 Lamiaceae Salvia pratensis Czech Republic, South Moravia*,3 Lamiaceae Linaria vulgaris Czech Republic, South Moravia*,3 Plantaginaceae Plantago lanceolata Germany, Berghausen*,1, Czech Republic, South Plantaginaceae Moravia*,3 Holcus spec. Germany, Berghausen(*),1 Poaceae Rumex crispus Czech Republic, South Moravia*,3 Polygonaceae Ranunculus polyanthemus South Moravia*,3 Ranunculaceae Ranunculus spec. Germany, Berghausen1 Ranunculaceae Agrimonia eupatoria Germany, Berghausen*,1 Rosaceae Geum urbanum Germany, Berghausen*,1 Rosaceae Potentilla tabernaemontani Czech Republic, South Moravia*,3 Rosaceae Rubus spec. (stem, leaf) Germany, Berghausen*,1 Rosaceae 38

Galium spec. Germany, Berghausen*,1 Rubiaceae Galium verum Czech Republic, South Moravia*,3 Rubiaceae Salix alba Czech Republic, South Moravia*,3 Salicaceae Verbascum phoeniceum Czech Republic, South Moravia*,3 Scrophulariaceae 1410 1411 1412 1413 1414 1415

1416 1417 Supplemental Figure 1. Growth of milkweed bug larvae on four different seed diets. (a-c) Weight gain of H. superbus, 1418 L. equestris, and S. saxatilis larvae feeding on sunflower seeds (control, black), a non-toxic seed mixture (green), a 1419 seed mixture with toxic seeds (red), or toxic seeds only (blue). Points are average weights of 1-3 larvae per petri dish 1420 (replicates), measured sequentially over three weeks. Growth was modelled as an asymptotic process using a non- 1421 linear mixed effects model (function nlme with SSasymp in R) with log-weight as the response, a random effect of 1422 petri dish to account for repeated measures, and fixed effects of species and treatment on individual model parameters 1423 (K: species × treatment, F6,374 = 165.4, p < 0.001; M0: species, F2,374 = 202.7, p < 0.001; log-rate constant: species, 1424 F2,374 = 83.7, p < 0.001 , treatment, F3,374 = 4.9, p = 0.002). Solid lines are model predictions, and shaded areas are 95% 1425 population prediction intervals, generated by drawing random values from the estimated sampling distribution of each 1426 regression parameter. (d-e) Absolute growth rates for H. superbus, L. equestris, and S. saxatilis larvae feeding on 1427 different seed diets. AGR values were calculated for growth on day 22 (final day of the experiment), using the model 1428 parameters from the asymptotic regression model displayed in panels a-c. Error bars are 95% population prediction 1429 intervals. 40

1430 1431 Supplemental Figure 2. Comparison of HPLC-chromatograms obtained from field-collected milkweed bugs with 1432 authentic reference compounds. (a) H. superbus from D. purpurea compared to compounds reported from D. 1433 purpurea: 1 = digoxigenin, 2 = lanatoside C, 3 = digoxin, 4 = gitoxigenin, 5 = purpurea glycoside B, 6 = purpurea 1434 glycoside A, 7 = gitoxin, 8 = digitoxigenin, 9 = digitoxin. (b) L. equestris collected from A. vernalis compared to 1 = 1435 k-strophanthosid, 2 = strophanthidin and 3 = cymarin that all occur in A. vernalis. (c) S. saxatilis from C. autumnale 1436 compared to authentic colchicum alkaloid standards: 1 = colchicoside, 2 = 3-demethyl colchicine, 3 = 2-demethyl 1437 colchicine, and 4 = colchicine. (d) H. superbus from E. crepidifolium compared to erysimoside (1). (e) S. pandurus 1438 from U. maritima compared to the Urginea bufadienolides 1 = scillaren A, 2 = proscillaridin A. 41

1439 1440 Supplemental Figure 3. Comparison of milkweed bug and host plant seed extracts with HPLC-DAD. (a) H. 1441 superbus vs. D. purpurea seeds, (b) L. equestris vs. A. vernalis seeds, (c) S. pandurus vs. U. maritima seeds, (d) H. 1442 superbus vs. seeds of E. crepidifolium, (e) S. saxatilis vs. C. autumnale seeds. Top chromatograms (red) always 1443 represent seed; bottom chromatograms (black) always represent insect extracts. Asterisks indicate the most 1444 prominent cardiac glycoside or colchicum alkaloid peaks, respectively. 42

1445 1446 Supplemental Figure 4. Concentration of colchicum alkaloids in Spilostethus saxatilis haemolymph (n = 12) and 1447 defensive secretion (n = 7). Diamonds are means ± SE, circles represent jittered raw data. Subset of six paired samples 1448 obtained from the same individuals used for statistical comparison (inset).

1449

1450

1451 1452 Supplemental Figure 5. Origin of S. saxatilis museum specimens used for chemical analysis. Numbers in 1453 parentheses indicate numbers of specimens sampled per location (MFNB: Museum für Naturkunde, Berlin, 1454 Germany; SMNK: Staatliches Museum für Naturkunde Karlsruhe, Germany; SDEI: Senckenberg Deutsches 1455 Entomologisches Institut, Müncheberg, Germany). Countries on labels were translated. 1: Germany, Odenwald, 1456 Dreieichenhain, September 15th 1925, MFNB (1); 2: Germany, Bamberg, Hallstadt, Börstig, August 3rd 1940, leg. 1457 Schneid, MFNB (1); 3: Germany, Baden, Bergstraße, Weinheim, May 1952, leg. H. Nowotny, SMNK (1); 4: 1458 Germany, Bienwald, Büchelberg, September 10th 1987, leg. Roesler, SMNK (2); 5: Germany, Karlsruhe, Maxau, 1459 August 22nd 1959, leg. Kormann; Germany, Karlsruhe, Maxau, August 1947, leg. Nowotny, SMNK (2); 6: Germany, 1460 Baden, Kaiserstuhl, June 21st 1953, leg. H. Nowotny, SMNK (1); 7: Germany, Baden, Hegau, August 8th-20th 1935, 1461 leg. Leininger, SMNK (2); 8: Germany, Bodensee, Allensbach, September 1919, leg. W. Ramme, MFNB (2); 9: 1462 Germany, Wollmatingen, August 10th 1928, leg. Leininger, SMNK (2); 10: Germany, Berchtesgaden, 1463 Bischofswiesen, August 21st-28th 1958, leg. Papperitz, SDEI (1); 11: Switzerland, Glarisegg, May 6th-10th 1906, 1464 SDEI (1); 12: Hungary, Mecsek-Gebirge, Mánfa, May 30th 1977, leg. U. Göllner, MFNB (1); 13: Romania, Brașov, 1465 August 13th 1905, leg. E.J. Lehmann, MFNB (1); 14: Italy, Alpi Marittime Natural Park, Juniperus phoenicea 1466 Riserva Naturale, August 10th 2013, leg. J. Deckert, MFNB (2); 15: Bosnia and Herzegovina, Bjelašnica, July 21st 1467 and 23rd 1909, leg. F. Schumacher, MFNB (2); 16: France, Riez, July 1985, leg. J. Haupt, MFNB (1); 17: Croatia, 1468 Dubrovnik, April 13th and May 1st 1938, leg. Dr. Feige, SDEI (3); 18: Albania, Dajti Südhang, June 30th 1961, leg. 1469 Expedition DEI, SDEI (1); 19: Republic of Macedonia, Hügel bei Stari Dojran, August 3rd 1975, leg. U. Göllner, 1470 MFNB (1); 20: France, Corsica, Bocognano, 1905, leg. O. Leonhard, SDEI (1); 21: Kingdom of Morocco, Taza, 1471 MFNB (1).

1472

1473 1474 Supplemental Figure 6. Natural history observations on milkweed bug species. (a) Eggs of S. saxatilis (white 1475 arrow) in a C. autumnale infructescence (Germany, Baden-Württemberg, Berghausen, June 1st, 2016). (b) Early 1476 instar larvae of S. saxatilis sitting on a dry Colchicum seedpod (Germany, Baden-Württemberg, Berghausen, June 1477 15th, 2016). (c) Adults of S. saxatilis feeding on a C. autumnale flower (Germany, Baden-Württemberg, Berghausen, 1478 September 18th, 2015). (d) H. superbus on D. purpurea (Germany, Baden-Württemberg, Eberbach, June 5th, 2016). 1479 (e) S. pandurus on U. maritima (Spain, Sierra de Aracena and Picos de Aroche Natural Park, Cañaveral de León, 1480 August 20th, 2014). (f) H. superbus on E. crepidifolium (Germany, Rheinland-Pfalz, Schloßböckelheim, June 13th, 1481 2016). (g) L. equestris on A. vernalis (Germany, Brandenburg, Mallnow, April 19th, 2016. 45

1482 1483 Supplemental Figure 7. Consumption of four species of milkweed bugs by lacewing larvae (C. carnea). Early instar 1484 larvae of milkweed bugs were raised either on sunflower seeds (controls) or on seeds of plant species containing 1485 toxins for sequestration. Left panel: Body mass of milkweed bug larvae before (open bars) and after lacewing attacks 1486 (hatched bars). (a) H. superbus on D. purpurea, (b) L. equestris on A. vernalis, (c) S. pandurus on U. maritima, (d) S. 1487 saxatilis on C. autumnale. Right panel: Body mass of lacewing larvae before (open bars) and after feeding on 1488 milkweed bug larvae (hatched bars). (e) H. superbus on D. purpurea, (f) L. equestris on A. vernalis, (g) S. pandurus 1489 on U. maritima, (h) S. saxatilis on C. autumnale. Shown are means ± SE. Asterisks indicate significant differences at 1490 p = 0.05, n.s. = not significant. 46

1491 1492 Supplemental Figure 8. Attack latencies of juvenile great tits (Parus major) preying on milkweed bugs and control 1493 palatable prey. (A) Attack latencies in first encounters with novel palatable prey (crickets; hatched rectangles) and 1494 milkweed bugs (H. superbus, L. equestris, and S. saxatilis; open rectangles). (B) Changes in attack latencies across 1495 three successive trials with milkweed bugs raised either on seeds from toxic host plants or sunflower seeds as a 1496 control (data from H. superbus, L. equestris, and S. saxatilis pooled). Only the data from bugs that were actually 1497 attacked by birds in the respective trials are included. Boxes represent median (points), quartiles (rectangles) and 1498 range (whiskers). 47

1499 1500 Supplemental Figure 9. Durations of discomfort-indicating behaviors (beak wiping, head shaking) exhibited by 1501 juvenile great tits (Parus major) following their first contact (attack and handling) with bugs either raised on seeds of 1502 toxic host plants or sunflower as a control. a, b, c: birds tested with H. superbus, L. equestris, and S. saxatilis. Boxes 1503 represent median (points), quartiles (rectangles) and range (whiskers).

1504 1505 Supplemental Figure 10. Resistance of P. apterus, O. fasciatus, S. saxatilis, and S. pandurus to injected ouabain 1506 and colchicine. (a) Effect of increasing doses of injected colchicine and ouabain (d) on P. apterus, (b) Effect of 1507 increasing doses of injected colchicine on O. fasciatus, (c) Effect of increasing doses of injected colchicine on S. 1508 pandurus. (e) Effect of injected ouabain and a high dose of colchicine on S. saxatilis. Bars show proportions of 1509 individuals that either showed signs of intoxication (open) or showed no signs of intoxication (hatched). Numbers in 1510 stacked bars represent the actual number of specimens. 49